Use BF16 for llama v3 by default.

* add support for l3b, new tokenizer

* add todo

* Add todo and use k_s model

* Use the official tokenizers.

---------

Co-authored-by: laurent <laurent.mazare@gmail.com>

Cargo.toml

@@ -9,6 +9,7 @@ members = [
     "candle-transformers",
     "candle-wasm-examples/*",
     "candle-wasm-tests",
+    "tensor-tools",
 ]
 exclude = [
    "candle-flash-attn",
@@ -19,7 +20,7 @@ exclude = [
 resolver = "2"
 
 [workspace.package]
-version = "0.4.2"
+version = "0.5.0"
 edition = "2021"
 description = "Minimalist ML framework."
 repository = "https://github.com/huggingface/candle"
@@ -28,17 +29,18 @@ categories = ["science"]
 license = "MIT OR Apache-2.0"
 
 [workspace.dependencies]
+ab_glyph = "0.2.23"
 accelerate-src = { version = "0.3.2" }
 anyhow = { version = "1", features = ["backtrace"] }
 byteorder = "1.4.3"
-candle = { path = "./candle-core", package = "candle-core", version = "0.4.2" }
-candle-datasets = { path = "./candle-datasets", version = "0.4.2" }
-candle-flash-attn = { path = "./candle-flash-attn", version = "0.4.2" }
-candle-kernels = { path = "./candle-kernels", version = "0.4.2" }
-candle-metal-kernels = { path = "./candle-metal-kernels", version = "0.4.2" }
-candle-nn = { path = "./candle-nn", version = "0.4.2" }
-candle-onnx = { path = "./candle-onnx", version = "0.4.2" }
-candle-transformers = { path = "./candle-transformers", version = "0.4.2" }
+candle = { path = "./candle-core", package = "candle-core", version = "0.5.0" }
+candle-datasets = { path = "./candle-datasets", version = "0.5.0" }
+candle-flash-attn = { path = "./candle-flash-attn", version = "0.5.0" }
+candle-kernels = { path = "./candle-kernels", version = "0.5.0" }
+candle-metal-kernels = { path = "./candle-metal-kernels", version = "0.5.0" }
+candle-nn = { path = "./candle-nn", version = "0.5.0" }
+candle-onnx = { path = "./candle-onnx", version = "0.5.0" }
+candle-transformers = { path = "./candle-transformers", version = "0.5.0" }
 clap = { version = "4.2.4", features = ["derive"] }
 criterion = { version = "0.5.1", default-features=false }
 cudarc = { version = "0.10.0", features = ["f16"] }
@@ -46,19 +48,18 @@ fancy-regex = "0.13.0"
 gemm = { version = "0.17.0", features = ["wasm-simd128-enable"] }
 hf-hub = "0.3.0"
 half = { version = "2.3.1", features = ["num-traits", "use-intrinsics", "rand_distr"] }
-image = { version = "0.24.7", default-features = false, features = ["jpeg", "png"] }
-imageproc = { version = "0.23.0", default-features = false }
+image = { version = "0.25.0", default-features = false, features = ["jpeg", "png"] }
+imageproc = { version = "0.24.0", default-features = false }
 intel-mkl-src = { version = "0.8.1", features = ["mkl-static-lp64-iomp"] }
 libc = { version = "0.2.147" }
 log = "0.4"
 memmap2 = { version = "0.9.3", features = ["stable_deref_trait"] }
 num_cpus = "1.15.0"
 num-traits = "0.2.15"
-parquet = { version = "50.0.0" }
+parquet = { version = "51.0.0" }
 rand = "0.8.5"
 rand_distr = "0.4.3"
 rayon = "1.7.0"
-rusttype = { version = "0.9", default-features = false }
 safetensors = "0.4.1"
 serde = { version = "1.0.171", features = ["derive"] }
 serde_plain = "1.0.2"

README.md

@@ -63,8 +63,9 @@ We also provide a some command line based examples using state of the art models
 - [LLaMA and LLaMA-v2](./candle-examples/examples/llama/): general LLM, includes
   the SOLAR-10.7B variant.
 - [Falcon](./candle-examples/examples/falcon/): general LLM.
-- [Gemma](./candle-examples/examples/gemma/): 2b and 7b general LLMs from Google
-  Deepmind.
+- [Gemma](./candle-examples/examples/gemma/): 2b and 7b general LLMs from Google Deepmind.
+- [RecurrentGemma](./candle-examples/examples/recurrent-gemma/): 2b and 7b
+  Griffin based models from Google that mix attention with a RNN like state.
 - [Phi-1, Phi-1.5, and Phi-2](./candle-examples/examples/phi/): 1.3b and 2.7b general LLMs with performance on par with LLaMA-v2 7b.
 - [StableLM-3B-4E1T](./candle-examples/examples/stable-lm/): a 3b general LLM
   pre-trained on 1T tokens of English and code datasets. Also supports
@@ -125,10 +126,14 @@ We also provide a some command line based examples using state of the art models
   [RepVGG](./candle-examples/examples/repvgg): computer vision models.
 - [BLIP](./candle-examples/examples/blip/): image to text model, can be used to
   generate captions for an image.
+- [CLIP](./candle-examples/examples/clip/): multi-model vision and language
+  model.
 - [TrOCR](./candle-examples/examples/trocr/): a transformer OCR model, with
   dedicated submodels for hand-writing and printed recognition.
 - [Marian-MT](./candle-examples/examples/marian-mt/): neural machine translation
   model, generates the translated text from the input text.
+- [Moondream](./candle-examples/examples/moondream/): tiny computer-vision model 
+  that can answer real-world questions about images.
 
 Run them using commands like:
 ```
@@ -172,6 +177,7 @@ And then head over to
 - [`candle-vllm`](https://github.com/EricLBuehler/candle-vllm): Efficient platform for inference and
   serving local LLMs including an OpenAI compatible API server.
 - [`candle-ext`](https://github.com/mokeyish/candle-ext): An extension library to Candle that provides PyTorch functions not currently available in Candle.
+- [`candle-coursera-ml`](https://github.com/vishpat/candle-coursera-ml): Implementation of ML algorithms from Coursera's [Machine Learning Specialization](https://www.coursera.org/specializations/machine-learning-introduction) course.
 - [`kalosm`](https://github.com/floneum/floneum/tree/master/interfaces/kalosm): A multi-modal meta-framework in Rust for interfacing with local pre-trained models with support for controlled generation, custom samplers, in-memory vector databases, audio transcription, and more.
 - [`candle-sampling`](https://github.com/EricLBuehler/candle-sampling): Sampling techniques for Candle.
 - [`gpt-from-scratch-rs`](https://github.com/jeroenvlek/gpt-from-scratch-rs): A port of Andrej Karpathy's _Let's build GPT_ tutorial on YouTube showcasing the Candle API on a toy problem.
@@ -206,7 +212,7 @@ If you have an addition to this list, please submit a pull request.
         - Replit-code-v1.5-3B.
         - Bert.
         - Yi-6B and Yi-34B.
-        - Qwen1.5.
+        - Qwen1.5, Qwen1.5 MoE.
         - RWKV v5 and v6.
     - Quantized LLMs.
         - Llama 7b, 13b, 70b, as well as the chat and code variants.
@@ -369,9 +375,9 @@ git submodule update --init
 /usr/include/c++/11/bits/std_function.h:530:146: error: parameter packs not expanded with ‘...’:
 ```
 
-This is a bug in gcc-11 triggered by the Cuda compiler. To fix this, install a different, supported gcc version - for example gcc-10, and specify the path to the compiler in the CANDLE_NVCC_CCBIN environment variable.
+This is a bug in gcc-11 triggered by the Cuda compiler. To fix this, install a different, supported gcc version - for example gcc-10, and specify the path to the compiler in the NVCC_CCBIN environment variable.
 ```
-env CANDLE_NVCC_CCBIN=/usr/lib/gcc/x86_64-linux-gnu/10 cargo ...
+env NVCC_CCBIN=/usr/lib/gcc/x86_64-linux-gnu/10 cargo ...
 ```
 
 #### Linking error on windows when running rustdoc or mdbook tests

candle-core/benches/bench_main.rs

@@ -5,5 +5,7 @@ criterion_main!(
     benchmarks::affine::benches,
     benchmarks::matmul::benches,
     benchmarks::random::benches,
-    benchmarks::where_cond::benches
+    benchmarks::where_cond::benches,
+    benchmarks::conv_transpose2d::benches,
+    benchmarks::qmatmul::benches,
 );

candle-core/benches/benchmarks/conv_transpose2d.rs

@@ -0,0 +1,59 @@
+use crate::benchmarks::{BenchDevice, BenchDeviceHandler};
+use candle_core::{DType, Device, Tensor};
+use criterion::{black_box, criterion_group, Criterion, Throughput};
+use std::time::Instant;
+
+fn run(
+    x: &Tensor,
+    k: &Tensor,
+    padding: usize,
+    output_padding: usize,
+    stride: usize,
+    dilation: usize,
+) {
+    x.conv_transpose2d(k, padding, output_padding, stride, dilation)
+        .unwrap();
+}
+
+fn run_benchmark(c: &mut Criterion, device: &Device, dtype: DType, name: &str) {
+    let t = Tensor::arange(0.0f32, 10000.0, device)
+        .unwrap()
+        .reshape((1, 4, 50, 50))
+        .unwrap()
+        .to_dtype(dtype)
+        .unwrap();
+
+    let kernel = Tensor::arange(0.0f32, 100.0, device)
+        .unwrap()
+        .reshape((4, 1, 5, 5))
+        .unwrap()
+        .to_dtype(dtype)
+        .unwrap();
+
+    let flops = t.dims().iter().product::<usize>() * dtype.size_in_bytes();
+
+    let mut group = c.benchmark_group(device.bench_name(name));
+    group.throughput(Throughput::Bytes(flops as u64));
+    group.bench_function("iter", move |b| {
+        b.iter_custom(|iters| {
+            let start = Instant::now();
+            for _i in 0..iters {
+                run(black_box(&t), black_box(&kernel), 1, 0, 1, 2);
+            }
+            device.sync().unwrap();
+            start.elapsed()
+        })
+    });
+    group.finish();
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let handler = BenchDeviceHandler::new().unwrap();
+    for device in handler.devices {
+        run_benchmark(c, &device, DType::F32, "conv_transpose2d_f32");
+        run_benchmark(c, &device, DType::F16, "conv_transpose2d_f16");
+        run_benchmark(c, &device, DType::BF16, "conv_transpose2d_bf16");
+    }
+}
+
+criterion_group!(benches, criterion_benchmark);

candle-core/benches/benchmarks/mod.rs

@@ -1,5 +1,7 @@
 pub(crate) mod affine;
+pub(crate) mod conv_transpose2d;
 pub(crate) mod matmul;
+pub(crate) mod qmatmul;
 pub(crate) mod random;
 pub(crate) mod where_cond;
 

candle-core/benches/benchmarks/qmatmul.rs

@@ -0,0 +1,72 @@
+use crate::benchmarks::{BenchDevice, BenchDeviceHandler};
+use candle_core::{
+    quantized::{self, GgmlDType, QMatMul},
+    Device, Module, Tensor,
+};
+use criterion::{black_box, criterion_group, Criterion, Throughput};
+use std::time::Instant;
+
+fn run(matmul: &QMatMul, x: &Tensor) {
+    matmul.forward(&x).unwrap();
+}
+
+fn run_bench(c: &mut Criterion, device: &Device, dtype: GgmlDType) {
+    let b = 1;
+    let m = 1;
+    let n = 1024;
+    let k = 1024;
+
+    let lhs = (0..(m * k))
+        .map(|v| v as f32 / (m * k) as f32)
+        .collect::<Vec<_>>();
+    let rhs = (0..(k * n))
+        .map(|v| v as f32 / (n * k) as f32)
+        .collect::<Vec<_>>();
+
+    let lhs = Tensor::from_slice(&lhs, (m, k), device).unwrap();
+    let rhs = Tensor::from_slice(&rhs, (k, n), device).unwrap();
+
+    let qtensor = quantized::QTensor::quantize(&rhs.t().unwrap(), dtype).unwrap();
+    let matmul = quantized::QMatMul::from_qtensor(qtensor).unwrap();
+
+    let flops = b * m * n * k;
+
+    let mut group = c.benchmark_group(device.bench_name(format!("qmatmul_{:?}", dtype)));
+    group.sample_size(200);
+    group.throughput(Throughput::Bytes(flops as u64));
+    group.bench_function("iter", move |b| {
+        b.iter_custom(|iters| {
+            let start = Instant::now();
+            for _i in 0..iters {
+                run(black_box(&matmul), black_box(&lhs));
+            }
+            device.sync().unwrap();
+            start.elapsed()
+        })
+    });
+    group.finish();
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let handler = BenchDeviceHandler::new().unwrap();
+    for device in handler.devices {
+        for dtype in vec![
+            GgmlDType::F32,
+            GgmlDType::F16,
+            GgmlDType::Q4_0,
+            GgmlDType::Q4_1,
+            GgmlDType::Q5_0,
+            GgmlDType::Q5_1,
+            GgmlDType::Q8_0,
+            GgmlDType::Q2K,
+            GgmlDType::Q3K,
+            GgmlDType::Q4K,
+            GgmlDType::Q5K,
+            GgmlDType::Q6K,
+        ] {
+            run_bench(c, &device, dtype);
+        }
+    }
+}
+
+criterion_group!(benches, criterion_benchmark);

candle-core/src/backend.rs

@@ -127,11 +127,22 @@ pub trait BackendDevice: Sized + std::fmt::Debug + Clone {
 
     fn ones_impl(&self, _shape: &Shape, _dtype: DType) -> Result<Self::Storage>;
 
+    /// # Safety
+    /// This function is unsafe as it doesn't initialize the underlying data store.
+    /// The caller should ensure that the data is properly initialized as early as possible
+    /// after this call.
+    unsafe fn alloc_uninit(&self, _shape: &Shape, _dtype: DType) -> Result<Self::Storage>;
+
     fn storage_from_cpu_storage(&self, _: &CpuStorage) -> Result<Self::Storage>;
 
+    fn storage_from_cpu_storage_owned(&self, _: CpuStorage) -> Result<Self::Storage>;
+
     fn rand_uniform(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage>;
 
     fn rand_normal(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage>;
 
     fn set_seed(&self, _: u64) -> Result<()>;
+
+    /// Synchronize should block until all the operations on the device are completed.
+    fn synchronize(&self) -> Result<()>;
 }

candle-core/src/backprop.rs

@@ -1,3 +1,4 @@
+/// Methods for backpropagation of gradients.
 use crate::op::{BinaryOp, Op, ReduceOp, UnaryOp};
 use crate::{Error, Result, Tensor, TensorId};
 use std::collections::HashMap;
@@ -111,7 +112,8 @@ impl Tensor {
                     }
                     Op::Unary(_node, UnaryOp::Ceil)
                     | Op::Unary(_node, UnaryOp::Floor)
-                    | Op::Unary(_node, UnaryOp::Round) => nodes,
+                    | Op::Unary(_node, UnaryOp::Round)
+                    | Op::Unary(_node, UnaryOp::Sign) => nodes,
                     Op::Reshape(node)
                     | Op::UpsampleNearest1D { arg: node, .. }
                     | Op::UpsampleNearest2D { arg: node, .. }
@@ -310,9 +312,32 @@ impl Tensor {
                     Op::ConvTranspose1D { .. } => Err(Error::BackwardNotSupported {
                         op: "conv-transpose1d",
                     })?,
-                    Op::ConvTranspose2D { .. } => Err(Error::BackwardNotSupported {
-                        op: "conv-transpose2d",
-                    })?,
+                    Op::ConvTranspose2D {
+                        arg,
+                        kernel,
+                        padding,
+                        stride,
+                        dilation,
+                        output_padding: _output_padding,
+                    } => {
+                        let grad_arg = grad.conv2d(kernel, *padding, *dilation, *stride, 1)?;
+                        let sum_grad = grads.or_insert(arg)?;
+                        *sum_grad = sum_grad.add(&grad_arg)?;
+
+                        let grad_kernel = grad
+                            .transpose(0, 1)?
+                            .conv2d(&arg.transpose(0, 1)?, *padding, *stride, *dilation, 1)?
+                            .transpose(0, 1)?;
+                        let sum_grad = grads.or_insert(kernel)?;
+                        let (_, _, k0, k1) = kernel.dims4()?;
+                        let (_, _, g_k0, g_k1) = grad_kernel.dims4()?;
+                        let grad_kernel = if g_k0 != k0 || g_k1 != k1 {
+                            grad_kernel.narrow(2, 0, k0)?.narrow(3, 0, k1)?
+                        } else {
+                            grad_kernel
+                        };
+                        *sum_grad = sum_grad.add(&grad_kernel)?;
+                    }
                     Op::AvgPool2D {
                         arg,
                         kernel_size,
@@ -464,7 +489,6 @@ impl Tensor {
                         let sum_grad = grads.or_insert(arg)?;
                         *sum_grad = sum_grad.add(&grad)?;
                     }
-                    Op::Cmp(_args, _) => {}
                     Op::Reduce(arg, ReduceOp::Max, reduced_dims) => {
                         let node = broadcast_back(arg, node, reduced_dims)?;
                         let grad = broadcast_back(arg, &grad, reduced_dims)?;
@@ -554,20 +578,18 @@ impl Tensor {
                         let sum_grad = grads.or_insert(arg)?;
                         *sum_grad = sum_grad.add(&arg_grad)?
                     }
-                    Op::Reduce(_, ReduceOp::ArgMin, _) => {}
-                    Op::Reduce(_, ReduceOp::ArgMax, _) => {}
+                    Op::Unary(_, UnaryOp::Floor)
+                    | Op::Unary(_, UnaryOp::Round)
+                    | Op::Reduce(_, ReduceOp::ArgMin, _)
+                    | Op::Reduce(_, ReduceOp::ArgMax, _)
+                    | Op::Unary(_, UnaryOp::Sign)
+                    | Op::Cmp(_, _) => {}
                     Op::Reshape(arg) => {
                         let arg_grad = grad.reshape(arg.dims())?;
                         let sum_grad = grads.or_insert(arg)?;
                         *sum_grad = sum_grad.add(&arg_grad)?
                     }
                     Op::Unary(_, UnaryOp::Ceil) => Err(Error::BackwardNotSupported { op: "ceil" })?,
-                    Op::Unary(_, UnaryOp::Floor) => {
-                        Err(Error::BackwardNotSupported { op: "floor" })?
-                    }
-                    Op::Unary(_, UnaryOp::Round) => {
-                        Err(Error::BackwardNotSupported { op: "round" })?
-                    }
                     Op::Unary(arg, UnaryOp::Gelu) => {
                         let sum_grad = grads.or_insert(arg)?;
                         let cube = arg.powf(3.)?;
@@ -602,7 +624,7 @@ impl Tensor {
                     Op::Unary(arg, UnaryOp::Silu) => {
                         let sum_grad = grads.or_insert(arg)?;
                         // d/dx silu = sigmoid(x) * (1 + x * (1 - sigmoid(x)))
-                        let sigmoid_arg = (*node / arg)?;
+                        let sigmoid_arg = (arg.neg()?.exp()? + 1.)?.recip()?;
                         let silu_grad = (&sigmoid_arg * (1. + (arg * (1. - &sigmoid_arg)?)?)?)?;
                         *sum_grad = sum_grad.add(&(&grad * silu_grad)?)?
                     }
@@ -690,30 +712,38 @@ impl Tensor {
     }
 }
 
+/// A store for gradients, associating a tensor id to the corresponding gradient tensor, used for back propagation.
 #[derive(Debug)]
 pub struct GradStore(HashMap<TensorId, Tensor>);
 
 impl GradStore {
+    /// Create a new gradient store
     fn new() -> Self {
         GradStore(HashMap::new())
     }
 
+    /// Get the gradient tensor corresponding to the given tensor id
     pub fn get_id(&self, id: TensorId) -> Option<&Tensor> {
         self.0.get(&id)
     }
 
+    /// Get the gradient tensor associated with the given tensor
     pub fn get(&self, tensor: &Tensor) -> Option<&Tensor> {
         self.0.get(&tensor.id())
     }
 
+    /// Remove the gradient tensor associated with the given tensor, returning it if it exists
     pub fn remove(&mut self, tensor: &Tensor) -> Option<Tensor> {
         self.0.remove(&tensor.id())
     }
 
+    /// Insert a gradient tensor associated with the given tensor, returning the previous gradient tensor if it existed
     pub fn insert(&mut self, tensor: &Tensor, grad: Tensor) -> Option<Tensor> {
         self.0.insert(tensor.id(), grad)
     }
 
+    /// Get the gradient tensor associated with the given tensor, or, if it does not exist,
+    /// insert a tensor of zeroes, with the same shape and type as the given tensors and return it
     fn or_insert(&mut self, tensor: &Tensor) -> Result<&mut Tensor> {
         use std::collections::hash_map::Entry;
         let grad = match self.0.entry(tensor.id()) {

candle-core/src/cpu_backend/mod.rs

@@ -4,6 +4,11 @@ use crate::{DType, Error, IntDType, Layout, Result, Shape, WithDType};
 use half::{bf16, f16};
 use rayon::prelude::*;
 
+mod utils;
+pub use utils::{
+    binary_map, binary_map_vec, unary_map, unary_map_vec, Map1, Map1Any, Map2, Map2U8,
+};
+
 const USE_IM2COL_CONV1D: bool = true;
 const USE_IM2COL_CONV1D_TR: bool = true;
 const USE_IM2COL_CONV2D: bool = true;
@@ -24,102 +29,6 @@ pub enum CpuStorage {
 #[derive(Debug, Clone)]
 pub struct CpuDevice;
 
-pub trait Map1 {
-    fn f<T: WithDType>(&self, vs: &[T], layout: &Layout) -> Result<Vec<T>>;
-
-    fn map(&self, vs: &CpuStorage, layout: &Layout) -> Result<CpuStorage> {
-        match vs {
-            CpuStorage::U8(vs) => Ok(CpuStorage::U8(self.f(vs, layout)?)),
-            CpuStorage::U32(vs) => Ok(CpuStorage::U32(self.f(vs, layout)?)),
-            CpuStorage::I64(vs) => Ok(CpuStorage::I64(self.f(vs, layout)?)),
-            CpuStorage::BF16(vs) => Ok(CpuStorage::BF16(self.f(vs, layout)?)),
-            CpuStorage::F16(vs) => Ok(CpuStorage::F16(self.f(vs, layout)?)),
-            CpuStorage::F32(vs) => Ok(CpuStorage::F32(self.f(vs, layout)?)),
-            CpuStorage::F64(vs) => Ok(CpuStorage::F64(self.f(vs, layout)?)),
-        }
-    }
-}
-
-pub trait Map1Any {
-    fn f<T: WithDType, W: Fn(Vec<T>) -> CpuStorage>(
-        &self,
-        vs: &[T],
-        layout: &Layout,
-        wrap: W,
-    ) -> Result<CpuStorage>;
-
-    fn map(&self, vs: &CpuStorage, layout: &Layout) -> Result<CpuStorage> {
-        match vs {
-            CpuStorage::U8(vs) => Ok(self.f(vs, layout, CpuStorage::U8)?),
-            CpuStorage::U32(vs) => Ok(self.f(vs, layout, CpuStorage::U32)?),
-            CpuStorage::I64(vs) => Ok(self.f(vs, layout, CpuStorage::I64)?),
-            CpuStorage::BF16(vs) => Ok(self.f(vs, layout, CpuStorage::BF16)?),
-            CpuStorage::F16(vs) => Ok(self.f(vs, layout, CpuStorage::F16)?),
-            CpuStorage::F32(vs) => Ok(self.f(vs, layout, CpuStorage::F32)?),
-            CpuStorage::F64(vs) => Ok(self.f(vs, layout, CpuStorage::F64)?),
-        }
-    }
-}
-
-type C = CpuStorage;
-pub trait Map2 {
-    const OP: &'static str;
-    fn f<T: WithDType>(&self, v1: &[T], l1: &Layout, v2: &[T], l2: &Layout) -> Result<Vec<T>>;
-
-    fn map(
-        &self,
-        v1: &CpuStorage,
-        l1: &Layout,
-        v2: &CpuStorage,
-        l2: &Layout,
-    ) -> Result<CpuStorage> {
-        match (v1, v2) {
-            (C::U8(v1), C::U8(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::U32(v1), C::U32(v2)) => Ok(C::U32(self.f(v1, l1, v2, l2)?)),
-            (C::I64(v1), C::I64(v2)) => Ok(C::I64(self.f(v1, l1, v2, l2)?)),
-            (C::BF16(v1), C::BF16(v2)) => Ok(C::BF16(self.f(v1, l1, v2, l2)?)),
-            (C::F16(v1), C::F16(v2)) => Ok(C::F16(self.f(v1, l1, v2, l2)?)),
-            (C::F32(v1), C::F32(v2)) => Ok(C::F32(self.f(v1, l1, v2, l2)?)),
-            (C::F64(v1), C::F64(v2)) => Ok(C::F64(self.f(v1, l1, v2, l2)?)),
-            _ => Err(Error::DTypeMismatchBinaryOp {
-                lhs: v1.dtype(),
-                rhs: v2.dtype(),
-                op: Self::OP,
-            }
-            .bt()),
-        }
-    }
-}
-
-pub trait Map2U8 {
-    const OP: &'static str;
-    fn f<T: WithDType>(&self, v1: &[T], l1: &Layout, v2: &[T], l2: &Layout) -> Result<Vec<u8>>;
-
-    fn map(
-        &self,
-        v1: &CpuStorage,
-        l1: &Layout,
-        v2: &CpuStorage,
-        l2: &Layout,
-    ) -> Result<CpuStorage> {
-        match (v1, v2) {
-            (C::U8(v1), C::U8(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::U32(v1), C::U32(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::I64(v1), C::I64(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::BF16(v1), C::BF16(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::F16(v1), C::F16(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::F32(v1), C::F32(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            (C::F64(v1), C::F64(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
-            _ => Err(Error::DTypeMismatchBinaryOp {
-                lhs: v1.dtype(),
-                rhs: v2.dtype(),
-                op: Self::OP,
-            }
-            .bt()),
-        }
-    }
-}
-
 struct Cmp(CmpOp);
 impl Map2U8 for Cmp {
     const OP: &'static str = "cmp";
@@ -366,275 +275,6 @@ impl<'a> Map1 for ReduceSum<'a> {
     }
 }
 
-pub fn unary_map<T: Copy, U: Copy, F: FnMut(T) -> U>(
-    vs: &[T],
-    layout: &Layout,
-    mut f: F,
-) -> Vec<U> {
-    match layout.strided_blocks() {
-        crate::StridedBlocks::SingleBlock { start_offset, len } => vs
-            [start_offset..start_offset + len]
-            .iter()
-            .map(|&v| f(v))
-            .collect(),
-        crate::StridedBlocks::MultipleBlocks {
-            block_start_index,
-            block_len,
-        } => {
-            let mut result = Vec::with_capacity(layout.shape().elem_count());
-            // Specialize the case where block_len is one to avoid the second loop.
-            if block_len == 1 {
-                for index in block_start_index {
-                    let v = unsafe { vs.get_unchecked(index) };
-                    result.push(f(*v))
-                }
-            } else {
-                for index in block_start_index {
-                    for offset in 0..block_len {
-                        let v = unsafe { vs.get_unchecked(index + offset) };
-                        result.push(f(*v))
-                    }
-                }
-            }
-            result
-        }
-    }
-}
-
-pub fn unary_map_vec<T: Copy, U: Copy, F: FnMut(T) -> U, FV: FnMut(&[T], &mut [U])>(
-    vs: &[T],
-    layout: &Layout,
-    mut f: F,
-    mut f_vec: FV,
-) -> Vec<U> {
-    match layout.strided_blocks() {
-        crate::StridedBlocks::SingleBlock { start_offset, len } => {
-            let mut ys: Vec<U> = Vec::with_capacity(len);
-            let ys_to_set = ys.spare_capacity_mut();
-            let ys_to_set = unsafe { std::mem::transmute::<_, &mut [U]>(ys_to_set) };
-            f_vec(&vs[start_offset..start_offset + len], ys_to_set);
-            // SAFETY: values are all set by f_vec.
-            unsafe { ys.set_len(len) };
-            ys
-        }
-        crate::StridedBlocks::MultipleBlocks {
-            block_start_index,
-            block_len,
-        } => {
-            let el_count = layout.shape().elem_count();
-            // Specialize the case where block_len is one to avoid the second loop.
-            if block_len == 1 {
-                let mut result = Vec::with_capacity(el_count);
-                for index in block_start_index {
-                    let v = unsafe { vs.get_unchecked(index) };
-                    result.push(f(*v))
-                }
-                result
-            } else {
-                let mut ys: Vec<U> = Vec::with_capacity(el_count);
-                let ys_to_set = ys.spare_capacity_mut();
-                let ys_to_set = unsafe { std::mem::transmute::<_, &mut [U]>(ys_to_set) };
-                let mut dst_index = 0;
-                for src_index in block_start_index {
-                    let vs = &vs[src_index..src_index + block_len];
-                    let ys = &mut ys_to_set[dst_index..dst_index + block_len];
-                    f_vec(vs, ys);
-                    dst_index += block_len;
-                }
-                // SAFETY: values are all set by f_vec.
-                unsafe { ys.set_len(el_count) };
-                ys
-            }
-        }
-    }
-}
-
-// This function maps over two strided index sequences.
-pub fn binary_map<T: Copy, U: Copy, F: FnMut(T, T) -> U>(
-    lhs_l: &Layout,
-    rhs_l: &Layout,
-    lhs: &[T],
-    rhs: &[T],
-    mut f: F,
-) -> Vec<U> {
-    match (lhs_l.contiguous_offsets(), rhs_l.contiguous_offsets()) {
-        (Some((o_l1, o_l2)), Some((o_r1, o_r2))) => lhs[o_l1..o_l2]
-            .iter()
-            .zip(rhs[o_r1..o_r2].iter())
-            .map(|(&l, &r)| f(l, r))
-            .collect(),
-        (Some((o_l1, o_l2)), None) => {
-            // TODO: Maybe we want to avoid going through the layout twice.
-            match rhs_l.offsets_b() {
-                Some(ob) => {
-                    let mut i_in_block = 0;
-                    let mut i_right_broadcast = 0;
-                    lhs[o_l1..o_l2]
-                        .iter()
-                        .map(|&l| {
-                            let r = unsafe { rhs.get_unchecked(i_in_block + ob.start) };
-                            i_right_broadcast += 1;
-                            if i_right_broadcast >= ob.right_broadcast {
-                                i_in_block += 1;
-                                i_right_broadcast = 0;
-                            }
-                            if i_in_block >= ob.len {
-                                i_in_block = 0
-                            }
-                            f(l, *r)
-                        })
-                        .collect()
-                }
-                None => lhs_l
-                    .strided_index()
-                    .zip(rhs_l.strided_index())
-                    .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
-                    .collect(),
-            }
-        }
-        (None, Some((o_r1, o_r2))) => {
-            // TODO: Maybe we want to avoid going through the layout twice.
-            match lhs_l.offsets_b() {
-                Some(ob) => {
-                    let mut i_in_block = 0;
-                    let mut i_right_broadcast = 0;
-                    rhs[o_r1..o_r2]
-                        .iter()
-                        .map(|&r| {
-                            let l = unsafe { lhs.get_unchecked(i_in_block + ob.start) };
-                            i_right_broadcast += 1;
-                            if i_right_broadcast >= ob.right_broadcast {
-                                i_in_block += 1;
-                                i_right_broadcast = 0;
-                            }
-                            if i_in_block >= ob.len {
-                                i_in_block = 0
-                            }
-                            f(*l, r)
-                        })
-                        .collect()
-                }
-                None => lhs_l
-                    .strided_index()
-                    .zip(rhs_l.strided_index())
-                    .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
-                    .collect(),
-            }
-        }
-        _ => lhs_l
-            .strided_index()
-            .zip(rhs_l.strided_index())
-            .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
-            .collect(),
-    }
-}
-
-// Similar to binary_map but with vectorized variants.
-pub fn binary_map_vec<T: Copy, F: FnMut(T, T) -> T, FV: FnMut(&[T], &[T], &mut [T])>(
-    lhs_l: &Layout,
-    rhs_l: &Layout,
-    lhs: &[T],
-    rhs: &[T],
-    mut f: F,
-    mut f_vec: FV,
-) -> Vec<T> {
-    let el_count = lhs_l.shape().elem_count();
-    match (lhs_l.contiguous_offsets(), rhs_l.contiguous_offsets()) {
-        (Some((o_l1, o_l2)), Some((o_r1, o_r2))) => {
-            let mut ys: Vec<T> = Vec::with_capacity(el_count);
-            let ys_to_set = ys.spare_capacity_mut();
-            let ys_to_set = unsafe { std::mem::transmute::<_, &mut [T]>(ys_to_set) };
-            f_vec(&lhs[o_l1..o_l2], &rhs[o_r1..o_r2], ys_to_set);
-            // SAFETY: values are all set by f_vec.
-            unsafe { ys.set_len(el_count) };
-            ys
-        }
-        (Some((o_l1, o_l2)), None) => match rhs_l.offsets_b() {
-            Some(ob) if ob.right_broadcast == 1 => {
-                let rhs = &rhs[ob.start..ob.start + ob.len];
-                let mut ys: Vec<T> = Vec::with_capacity(el_count);
-                let ys_to_set = ys.spare_capacity_mut();
-                let ys_to_set = unsafe { std::mem::transmute::<_, &mut [T]>(ys_to_set) };
-                let mut dst_i = 0;
-                for src_i in (o_l1..o_l2).step_by(ob.len) {
-                    f_vec(
-                        &lhs[src_i..src_i + ob.len],
-                        rhs,
-                        &mut ys_to_set[dst_i..dst_i + ob.len],
-                    );
-                    dst_i += ob.len;
-                }
-                // SAFETY: values are all set by f_vec.
-                unsafe { ys.set_len(el_count) };
-                ys
-            }
-            Some(ob) => {
-                let rhs = &rhs[ob.start..ob.start + ob.len];
-                let mut ys = lhs[o_l1..o_l2].to_vec();
-                for idx_l in 0..ob.left_broadcast {
-                    let start = idx_l * ob.len * ob.right_broadcast;
-                    for (i, &r) in rhs.iter().enumerate() {
-                        let start = start + i * ob.right_broadcast;
-                        for v in ys[start..start + ob.right_broadcast].iter_mut() {
-                            *v = f(*v, r)
-                        }
-                    }
-                }
-                ys
-            }
-            None => lhs_l
-                .strided_index()
-                .zip(rhs_l.strided_index())
-                .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
-                .collect(),
-        },
-        (None, Some((o_r1, o_r2))) => match lhs_l.offsets_b() {
-            Some(ob) if ob.right_broadcast == 1 => {
-                let lhs = &lhs[ob.start..ob.start + ob.len];
-                let mut ys: Vec<T> = Vec::with_capacity(el_count);
-                let ys_to_set = ys.spare_capacity_mut();
-                let ys_to_set = unsafe { std::mem::transmute::<_, &mut [T]>(ys_to_set) };
-                let mut dst_i = 0;
-                for src_i in (o_r1..o_r2).step_by(ob.len) {
-                    f_vec(
-                        lhs,
-                        &rhs[src_i..src_i + ob.len],
-                        &mut ys_to_set[dst_i..dst_i + ob.len],
-                    );
-                    dst_i += ob.len;
-                }
-                // SAFETY: values are all set by f_vec.
-                unsafe { ys.set_len(el_count) };
-                ys
-            }
-            Some(ob) => {
-                let lhs = &lhs[ob.start..ob.start + ob.len];
-                let mut ys = rhs[o_r1..o_r2].to_vec();
-                for idx_l in 0..ob.left_broadcast {
-                    let start = idx_l * ob.len * ob.right_broadcast;
-                    for (i, &l) in lhs.iter().enumerate() {
-                        let start = start + i * ob.right_broadcast;
-                        for v in ys[start..start + ob.right_broadcast].iter_mut() {
-                            *v = f(l, *v)
-                        }
-                    }
-                }
-                ys
-            }
-            None => lhs_l
-                .strided_index()
-                .zip(rhs_l.strided_index())
-                .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
-                .collect(),
-        },
-        _ => lhs_l
-            .strided_index()
-            .zip(rhs_l.strided_index())
-            .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
-            .collect(),
-    }
-}
-
 struct Affine(f64, f64);
 
 impl Map1 for Affine {
@@ -1564,6 +1204,30 @@ impl MatMul {
         }))
         .bt()
     }
+
+    fn ab_skip(&self, lhs_l: &Layout, rhs_l: &Layout) -> Result<(usize, usize)> {
+        let lhs_stride = lhs_l.stride();
+        let rhs_stride = rhs_l.stride();
+        let rank = lhs_stride.len();
+        let (_b, m, n, k) = self.0;
+        let a_skip: usize = match lhs_stride[..rank - 2] {
+            [s1, stride] if s1 == stride * lhs_l.dims()[1] => stride,
+            [_, stride] if lhs_l.dims()[0] == 1 => stride,
+            [stride, _] if lhs_l.dims()[1] == 1 => stride,
+            [stride] => stride,
+            [] => m * k,
+            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous lhs"))?,
+        };
+        let b_skip: usize = match rhs_stride[..rank - 2] {
+            [s1, stride] if s1 == stride * rhs_l.dims()[1] => stride,
+            [_, stride] if rhs_l.dims()[0] == 1 => stride,
+            [stride, _] if rhs_l.dims()[1] == 1 => stride,
+            [stride] => stride,
+            [] => n * k,
+            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous rhs"))?,
+        };
+        Ok((a_skip, b_skip))
+    }
 }
 
 impl Map2 for MatMul {
@@ -1597,18 +1261,7 @@ impl Map2 for MatMul {
         let rhs_cs = rhs_stride[rank - 1];
         let rhs_rs = rhs_stride[rank - 2];
 
-        let a_skip: usize = match lhs_stride[..rank - 2] {
-            [s1, stride] if s1 == stride * lhs_l.dims()[1] => stride,
-            [stride] => stride,
-            [] => m * k,
-            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous lhs"))?,
-        };
-        let b_skip: usize = match rhs_stride[..rank - 2] {
-            [s1, stride] if s1 == stride * rhs_l.dims()[1] => stride,
-            [stride] => stride,
-            [] => n * k,
-            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous rhs"))?,
-        };
+        let (a_skip, b_skip) = self.ab_skip(lhs_l, rhs_l)?;
         let c_skip: usize = m * n;
 
         let dst_shape: Shape = (m, n).into();
@@ -1668,20 +1321,8 @@ impl Map2 for MatMul {
 
         let lhs_stride = lhs_l.stride();
         let rhs_stride = rhs_l.stride();
-        let rank = lhs_stride.len();
 
-        let a_skip: usize = match lhs_stride[..rank - 2] {
-            [s1, stride] if s1 == stride * lhs_l.dims()[1] => stride,
-            [stride] => stride,
-            [] => m * k,
-            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous lhs"))?,
-        };
-        let b_skip: usize = match rhs_stride[..rank - 2] {
-            [s1, stride] if s1 == stride * rhs_l.dims()[1] => stride,
-            [stride] => stride,
-            [] => n * k,
-            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous rhs"))?,
-        };
+        let (a_skip, b_skip) = self.ab_skip(lhs_l, rhs_l)?;
         let c_skip: usize = m * n;
 
         let rhs_m1 = rhs_stride[rhs_stride.len() - 1];
@@ -1689,7 +1330,7 @@ impl Map2 for MatMul {
         let lhs_m1 = lhs_stride[lhs_stride.len() - 1];
         let lhs_m2 = lhs_stride[lhs_stride.len() - 2];
 
-        let (lda, transa) = if rhs_m1 == 1 && rhs_m2 == n {
+        let (lda, transa) = if (rhs_m1 == 1 || n == 1) && (rhs_m2 == n || k == 1) {
             (n as i32, b'N')
         } else if rhs_m1 == k && rhs_m2 == 1 {
             (k as i32, b'T')
@@ -1697,7 +1338,7 @@ impl Map2 for MatMul {
             Err(self.striding_error(lhs_l, rhs_l, "non-contiguous rhs"))?
         };
         // The b tensor has dims batching, m, k (lhs)
-        let (ldb, transb) = if lhs_m1 == 1 && lhs_m2 == k {
+        let (ldb, transb) = if (lhs_m1 == 1 || k == 1) && (lhs_m2 == k || m == 1) {
             (k as i32, b'N')
         } else if lhs_m1 == m && lhs_m2 == 1 {
             (m as i32, b'T')
@@ -1771,20 +1412,8 @@ impl Map2 for MatMul {
 
         let lhs_stride = lhs_l.stride();
         let rhs_stride = rhs_l.stride();
-        let rank = lhs_stride.len();
 
-        let a_skip: usize = match lhs_stride[..rank - 2] {
-            [s1, stride] if s1 == stride * lhs_l.dims()[1] => stride,
-            [stride] => stride,
-            [] => m * k,
-            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous lhs"))?,
-        };
-        let b_skip: usize = match rhs_stride[..rank - 2] {
-            [s1, stride] if s1 == stride * rhs_l.dims()[1] => stride,
-            [stride] => stride,
-            [] => n * k,
-            _ => Err(self.striding_error(lhs_l, rhs_l, "non-contiguous rhs"))?,
-        };
+        let (a_skip, b_skip) = self.ab_skip(lhs_l, rhs_l)?;
         let c_skip: usize = m * n;
 
         let rhs_m1 = rhs_stride[rhs_stride.len() - 1];
@@ -1792,7 +1421,7 @@ impl Map2 for MatMul {
         let lhs_m1 = lhs_stride[lhs_stride.len() - 1];
         let lhs_m2 = lhs_stride[lhs_stride.len() - 2];
 
-        let (lda, transa) = if rhs_m1 == 1 && rhs_m2 == n {
+        let (lda, transa) = if (rhs_m1 == 1 || n == 1) && (rhs_m2 == n || k == 1) {
             (n as i32, b'N')
         } else if rhs_m1 == k && rhs_m2 == 1 {
             (k as i32, b'T')
@@ -1800,7 +1429,7 @@ impl Map2 for MatMul {
             Err(self.striding_error(lhs_l, rhs_l, "non-contiguous rhs"))?
         };
         // The b tensor has dims batching, m, k (lhs)
-        let (ldb, transb) = if lhs_m1 == 1 && lhs_m2 == k {
+        let (ldb, transb) = if (lhs_m1 == 1 || k == 1) && (lhs_m2 == k || m == 1) {
             (k as i32, b'N')
         } else if lhs_m1 == m && lhs_m2 == 1 {
             (m as i32, b'T')
@@ -2582,7 +2211,10 @@ impl BackendStorage for CpuStorage {
             col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
         } else {
             // Make the kernel contiguous if not already the case.
-            let mut kernel_c = self.device().zeros_impl(kernel_l.shape(), kernel.dtype())?;
+            let mut kernel_c = unsafe {
+                self.device()
+                    .alloc_uninit(kernel_l.shape(), kernel.dtype())?
+            };
             kernel.copy_strided_src(&mut kernel_c, 0, kernel_l)?;
             let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
                 .transpose(1, 2)?
@@ -2590,7 +2222,7 @@ impl BackendStorage for CpuStorage {
             col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
         };
         let res_l = Layout::contiguous((b, l_out, params.c_out)).transpose(1, 2)?;
-        let mut res_t = self.device().zeros_impl(res_l.shape(), res.dtype())?;
+        let mut res_t = unsafe { self.device().alloc_uninit(res_l.shape(), res.dtype())? };
         res.copy_strided_src(&mut res_t, 0, &res_l)?;
         Ok(res_t)
     }
@@ -2681,7 +2313,10 @@ impl BackendStorage for CpuStorage {
             col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
         } else {
             // Make the kernel contiguous if not already the case.
-            let mut kernel_c = self.device().zeros_impl(kernel_l.shape(), kernel.dtype())?;
+            let mut kernel_c = unsafe {
+                self.device()
+                    .alloc_uninit(kernel_l.shape(), kernel.dtype())?
+            };
             kernel.copy_strided_src(&mut kernel_c, 0, kernel_l)?;
             let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
                 .transpose(1, 2)?
@@ -2691,7 +2326,7 @@ impl BackendStorage for CpuStorage {
         let res_l = Layout::contiguous((b, h_out, w_out, params.c_out))
             .transpose(1, 2)?
             .transpose(1, 3)?;
-        let mut res_t = self.device().zeros_impl(res_l.shape(), res.dtype())?;
+        let mut res_t = unsafe { self.device().alloc_uninit(res_l.shape(), res.dtype())? };
         res.copy_strided_src(&mut res_t, 0, &res_l)?;
         Ok(res_t)
     }
@@ -2814,6 +2449,10 @@ impl BackendDevice for CpuDevice {
         Ok(s.clone())
     }
 
+    fn storage_from_cpu_storage_owned(&self, s: CpuStorage) -> Result<Self::Storage> {
+        Ok(s)
+    }
+
     fn new(_: usize) -> Result<Self> {
         Ok(Self)
     }
@@ -2915,6 +2554,53 @@ impl BackendDevice for CpuDevice {
         }
     }
 
+    #[allow(clippy::uninit_vec)]
+    unsafe fn alloc_uninit(&self, shape: &Shape, dtype: DType) -> Result<CpuStorage> {
+        let elem_count = shape.elem_count();
+        // The code below is highly unsafe but hopefully not directly unsound as we only consider
+        // types that are Copy, not Drop, and for which all bit patterns are proper values.
+        // It's still pretty risky, see the following for more details:
+        // https://github.com/rust-lang/rust-clippy/issues/4483
+        let storage = match dtype {
+            DType::U8 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::U8(v)
+            }
+            DType::U32 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::U32(v)
+            }
+            DType::I64 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::I64(v)
+            }
+            DType::BF16 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::BF16(v)
+            }
+            DType::F16 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::F16(v)
+            }
+            DType::F32 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::F32(v)
+            }
+            DType::F64 => {
+                let mut v = Vec::with_capacity(elem_count);
+                v.set_len(elem_count);
+                CpuStorage::F64(v)
+            }
+        };
+        Ok(storage)
+    }
+
     fn ones_impl(&self, shape: &Shape, dtype: DType) -> Result<CpuStorage> {
         let elem_count = shape.elem_count();
         let storage = match dtype {
@@ -2942,6 +2628,10 @@ impl BackendDevice for CpuDevice {
         };
         Ok(storage)
     }
+
+    fn synchronize(&self) -> Result<()> {
+        Ok(())
+    }
 }
 
 #[macro_export]

candle-core/src/cpu_backend/utils.rs

@@ -0,0 +1,350 @@
+/// Helper functions to write CPU kernels.
+use crate::backend::BackendStorage;
+use crate::{Error, Layout, Result, WithDType};
+
+type C = super::CpuStorage;
+pub trait Map1 {
+    fn f<T: WithDType>(&self, vs: &[T], layout: &Layout) -> Result<Vec<T>>;
+
+    fn map(&self, vs: &C, layout: &Layout) -> Result<C> {
+        match vs {
+            C::U8(vs) => Ok(C::U8(self.f(vs, layout)?)),
+            C::U32(vs) => Ok(C::U32(self.f(vs, layout)?)),
+            C::I64(vs) => Ok(C::I64(self.f(vs, layout)?)),
+            C::BF16(vs) => Ok(C::BF16(self.f(vs, layout)?)),
+            C::F16(vs) => Ok(C::F16(self.f(vs, layout)?)),
+            C::F32(vs) => Ok(C::F32(self.f(vs, layout)?)),
+            C::F64(vs) => Ok(C::F64(self.f(vs, layout)?)),
+        }
+    }
+}
+
+pub trait Map1Any {
+    fn f<T: WithDType, W: Fn(Vec<T>) -> C>(&self, vs: &[T], layout: &Layout, wrap: W) -> Result<C>;
+
+    fn map(&self, vs: &C, layout: &Layout) -> Result<C> {
+        match vs {
+            C::U8(vs) => Ok(self.f(vs, layout, C::U8)?),
+            C::U32(vs) => Ok(self.f(vs, layout, C::U32)?),
+            C::I64(vs) => Ok(self.f(vs, layout, C::I64)?),
+            C::BF16(vs) => Ok(self.f(vs, layout, C::BF16)?),
+            C::F16(vs) => Ok(self.f(vs, layout, C::F16)?),
+            C::F32(vs) => Ok(self.f(vs, layout, C::F32)?),
+            C::F64(vs) => Ok(self.f(vs, layout, C::F64)?),
+        }
+    }
+}
+
+pub trait Map2 {
+    const OP: &'static str;
+    fn f<T: WithDType>(&self, v1: &[T], l1: &Layout, v2: &[T], l2: &Layout) -> Result<Vec<T>>;
+
+    fn map(&self, v1: &C, l1: &Layout, v2: &C, l2: &Layout) -> Result<C> {
+        match (v1, v2) {
+            (C::U8(v1), C::U8(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::U32(v1), C::U32(v2)) => Ok(C::U32(self.f(v1, l1, v2, l2)?)),
+            (C::I64(v1), C::I64(v2)) => Ok(C::I64(self.f(v1, l1, v2, l2)?)),
+            (C::BF16(v1), C::BF16(v2)) => Ok(C::BF16(self.f(v1, l1, v2, l2)?)),
+            (C::F16(v1), C::F16(v2)) => Ok(C::F16(self.f(v1, l1, v2, l2)?)),
+            (C::F32(v1), C::F32(v2)) => Ok(C::F32(self.f(v1, l1, v2, l2)?)),
+            (C::F64(v1), C::F64(v2)) => Ok(C::F64(self.f(v1, l1, v2, l2)?)),
+            _ => Err(Error::DTypeMismatchBinaryOp {
+                lhs: v1.dtype(),
+                rhs: v2.dtype(),
+                op: Self::OP,
+            }
+            .bt()),
+        }
+    }
+}
+
+pub trait Map2U8 {
+    const OP: &'static str;
+    fn f<T: WithDType>(&self, v1: &[T], l1: &Layout, v2: &[T], l2: &Layout) -> Result<Vec<u8>>;
+
+    fn map(&self, v1: &C, l1: &Layout, v2: &C, l2: &Layout) -> Result<C> {
+        match (v1, v2) {
+            (C::U8(v1), C::U8(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::U32(v1), C::U32(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::I64(v1), C::I64(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::BF16(v1), C::BF16(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::F16(v1), C::F16(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::F32(v1), C::F32(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            (C::F64(v1), C::F64(v2)) => Ok(C::U8(self.f(v1, l1, v2, l2)?)),
+            _ => Err(Error::DTypeMismatchBinaryOp {
+                lhs: v1.dtype(),
+                rhs: v2.dtype(),
+                op: Self::OP,
+            }
+            .bt()),
+        }
+    }
+}
+
+pub fn binary_map<T: Copy, U: Copy, F: FnMut(T, T) -> U>(
+    lhs_l: &Layout,
+    rhs_l: &Layout,
+    lhs: &[T],
+    rhs: &[T],
+    mut f: F,
+) -> Vec<U> {
+    match (lhs_l.contiguous_offsets(), rhs_l.contiguous_offsets()) {
+        (Some((o_l1, o_l2)), Some((o_r1, o_r2))) => lhs[o_l1..o_l2]
+            .iter()
+            .zip(rhs[o_r1..o_r2].iter())
+            .map(|(&l, &r)| f(l, r))
+            .collect(),
+        (Some((o_l1, o_l2)), None) => {
+            // TODO: Maybe we want to avoid going through the layout twice.
+            match rhs_l.offsets_b() {
+                Some(ob) => {
+                    let mut i_in_block = 0;
+                    let mut i_right_broadcast = 0;
+                    lhs[o_l1..o_l2]
+                        .iter()
+                        .map(|&l| {
+                            let r = unsafe { rhs.get_unchecked(i_in_block + ob.start) };
+                            i_right_broadcast += 1;
+                            if i_right_broadcast >= ob.right_broadcast {
+                                i_in_block += 1;
+                                i_right_broadcast = 0;
+                            }
+                            if i_in_block >= ob.len {
+                                i_in_block = 0
+                            }
+                            f(l, *r)
+                        })
+                        .collect()
+                }
+                None => lhs_l
+                    .strided_index()
+                    .zip(rhs_l.strided_index())
+                    .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
+                    .collect(),
+            }
+        }
+        (None, Some((o_r1, o_r2))) => {
+            // TODO: Maybe we want to avoid going through the layout twice.
+            match lhs_l.offsets_b() {
+                Some(ob) => {
+                    let mut i_in_block = 0;
+                    let mut i_right_broadcast = 0;
+                    rhs[o_r1..o_r2]
+                        .iter()
+                        .map(|&r| {
+                            let l = unsafe { lhs.get_unchecked(i_in_block + ob.start) };
+                            i_right_broadcast += 1;
+                            if i_right_broadcast >= ob.right_broadcast {
+                                i_in_block += 1;
+                                i_right_broadcast = 0;
+                            }
+                            if i_in_block >= ob.len {
+                                i_in_block = 0
+                            }
+                            f(*l, r)
+                        })
+                        .collect()
+                }
+                None => lhs_l
+                    .strided_index()
+                    .zip(rhs_l.strided_index())
+                    .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
+                    .collect(),
+            }
+        }
+        _ => lhs_l
+            .strided_index()
+            .zip(rhs_l.strided_index())
+            .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
+            .collect(),
+    }
+}
+
+// Similar to binary_map but with vectorized variants.
+pub fn binary_map_vec<T: Copy, F: FnMut(T, T) -> T, FV: FnMut(&[T], &[T], &mut [T])>(
+    lhs_l: &Layout,
+    rhs_l: &Layout,
+    lhs: &[T],
+    rhs: &[T],
+    mut f: F,
+    mut f_vec: FV,
+) -> Vec<T> {
+    let el_count = lhs_l.shape().elem_count();
+    match (lhs_l.contiguous_offsets(), rhs_l.contiguous_offsets()) {
+        (Some((o_l1, o_l2)), Some((o_r1, o_r2))) => {
+            let mut ys: Vec<T> = Vec::with_capacity(el_count);
+            let ys_to_set = ys.spare_capacity_mut();
+            let ys_to_set = unsafe { std::mem::transmute::<_, &mut [T]>(ys_to_set) };
+            f_vec(&lhs[o_l1..o_l2], &rhs[o_r1..o_r2], ys_to_set);
+            // SAFETY: values are all set by f_vec.
+            unsafe { ys.set_len(el_count) };
+            ys
+        }
+        (Some((o_l1, o_l2)), None) => match rhs_l.offsets_b() {
+            Some(ob) if ob.right_broadcast == 1 => {
+                let rhs = &rhs[ob.start..ob.start + ob.len];
+                let mut ys: Vec<T> = Vec::with_capacity(el_count);
+                let ys_to_set = ys.spare_capacity_mut();
+                let ys_to_set = unsafe { std::mem::transmute::<_, &mut [T]>(ys_to_set) };
+                let mut dst_i = 0;
+                for src_i in (o_l1..o_l2).step_by(ob.len) {
+                    f_vec(
+                        &lhs[src_i..src_i + ob.len],
+                        rhs,
+                        &mut ys_to_set[dst_i..dst_i + ob.len],
+                    );
+                    dst_i += ob.len;
+                }
+                // SAFETY: values are all set by f_vec.
+                unsafe { ys.set_len(el_count) };
+                ys
+            }
+            Some(ob) => {
+                let rhs = &rhs[ob.start..ob.start + ob.len];
+                let mut ys = lhs[o_l1..o_l2].to_vec();
+                for idx_l in 0..ob.left_broadcast {
+                    let start = idx_l * ob.len * ob.right_broadcast;
+                    for (i, &r) in rhs.iter().enumerate() {
+                        let start = start + i * ob.right_broadcast;
+                        for v in ys[start..start + ob.right_broadcast].iter_mut() {
+                            *v = f(*v, r)
+                        }
+                    }
+                }
+                ys
+            }
+            None => lhs_l
+                .strided_index()
+                .zip(rhs_l.strided_index())
+                .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
+                .collect(),
+        },
+        (None, Some((o_r1, o_r2))) => match lhs_l.offsets_b() {
+            Some(ob) if ob.right_broadcast == 1 => {
+                let lhs = &lhs[ob.start..ob.start + ob.len];
+                let mut ys: Vec<T> = Vec::with_capacity(el_count);
+                let ys_to_set = ys.spare_capacity_mut();
+                let ys_to_set = unsafe { std::mem::transmute::<_, &mut [T]>(ys_to_set) };
+                let mut dst_i = 0;
+                for src_i in (o_r1..o_r2).step_by(ob.len) {
+                    f_vec(
+                        lhs,
+                        &rhs[src_i..src_i + ob.len],
+                        &mut ys_to_set[dst_i..dst_i + ob.len],
+                    );
+                    dst_i += ob.len;
+                }
+                // SAFETY: values are all set by f_vec.
+                unsafe { ys.set_len(el_count) };
+                ys
+            }
+            Some(ob) => {
+                let lhs = &lhs[ob.start..ob.start + ob.len];
+                let mut ys = rhs[o_r1..o_r2].to_vec();
+                for idx_l in 0..ob.left_broadcast {
+                    let start = idx_l * ob.len * ob.right_broadcast;
+                    for (i, &l) in lhs.iter().enumerate() {
+                        let start = start + i * ob.right_broadcast;
+                        for v in ys[start..start + ob.right_broadcast].iter_mut() {
+                            *v = f(l, *v)
+                        }
+                    }
+                }
+                ys
+            }
+            None => lhs_l
+                .strided_index()
+                .zip(rhs_l.strided_index())
+                .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
+                .collect(),
+        },
+        _ => lhs_l
+            .strided_index()
+            .zip(rhs_l.strided_index())
+            .map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
+            .collect(),
+    }
+}
+
+pub fn unary_map<T: Copy, U: Copy, F: FnMut(T) -> U>(
+    vs: &[T],
+    layout: &Layout,
+    mut f: F,
+) -> Vec<U> {
+    match layout.strided_blocks() {
+        crate::StridedBlocks::SingleBlock { start_offset, len } => vs
+            [start_offset..start_offset + len]
+            .iter()
+            .map(|&v| f(v))
+            .collect(),
+        crate::StridedBlocks::MultipleBlocks {
+            block_start_index,
+            block_len,
+        } => {
+            let mut result = Vec::with_capacity(layout.shape().elem_count());
+            // Specialize the case where block_len is one to avoid the second loop.
+            if block_len == 1 {
+                for index in block_start_index {
+                    let v = unsafe { vs.get_unchecked(index) };
+                    result.push(f(*v))
+                }
+            } else {
+                for index in block_start_index {
+                    for offset in 0..block_len {
+                        let v = unsafe { vs.get_unchecked(index + offset) };
+                        result.push(f(*v))
+                    }
+                }
+            }
+            result
+        }
+    }
+}
+
+pub fn unary_map_vec<T: Copy, U: Copy, F: FnMut(T) -> U, FV: FnMut(&[T], &mut [U])>(
+    vs: &[T],
+    layout: &Layout,
+    mut f: F,
+    mut f_vec: FV,
+) -> Vec<U> {
+    match layout.strided_blocks() {
+        crate::StridedBlocks::SingleBlock { start_offset, len } => {
+            let mut ys: Vec<U> = Vec::with_capacity(len);
+            let ys_to_set = ys.spare_capacity_mut();
+            let ys_to_set = unsafe { std::mem::transmute::<_, &mut [U]>(ys_to_set) };
+            f_vec(&vs[start_offset..start_offset + len], ys_to_set);
+            // SAFETY: values are all set by f_vec.
+            unsafe { ys.set_len(len) };
+            ys
+        }
+        crate::StridedBlocks::MultipleBlocks {
+            block_start_index,
+            block_len,
+        } => {
+            let el_count = layout.shape().elem_count();
+            // Specialize the case where block_len is one to avoid the second loop.
+            if block_len == 1 {
+                let mut result = Vec::with_capacity(el_count);
+                for index in block_start_index {
+                    let v = unsafe { vs.get_unchecked(index) };
+                    result.push(f(*v))
+                }
+                result
+            } else {
+                let mut ys: Vec<U> = Vec::with_capacity(el_count);
+                let ys_to_set = ys.spare_capacity_mut();
+                let ys_to_set = unsafe { std::mem::transmute::<_, &mut [U]>(ys_to_set) };
+                let mut dst_index = 0;
+                for src_index in block_start_index {
+                    let vs = &vs[src_index..src_index + block_len];
+                    let ys = &mut ys_to_set[dst_index..dst_index + block_len];
+                    f_vec(vs, ys);
+                    dst_index += block_len;
+                }
+                // SAFETY: values are all set by f_vec.
+                unsafe { ys.set_len(el_count) };
+                ys
+            }
+        }
+    }
+}

candle-core/src/cuda_backend/device.rs

@@ -0,0 +1,415 @@
+use crate::backend::BackendDevice;
+use crate::{CpuStorage, DType, Layout, Result, Shape};
+pub use candle_kernels as kernels;
+pub use cudarc;
+use cudarc::driver::{CudaFunction, LaunchAsync, LaunchConfig};
+use half::{bf16, f16};
+use std::sync::{Arc, Mutex};
+
+use super::{CudaError, CudaStorage, CudaStorageSlice, WrapErr};
+
+/// Unique identifier for cuda devices.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub struct DeviceId(usize);
+
+impl DeviceId {
+    fn new() -> Self {
+        // https://users.rust-lang.org/t/idiomatic-rust-way-to-generate-unique-id/33805
+        use std::sync::atomic;
+        static COUNTER: atomic::AtomicUsize = atomic::AtomicUsize::new(1);
+        Self(COUNTER.fetch_add(1, atomic::Ordering::Relaxed))
+    }
+}
+
+struct CudaRng(cudarc::curand::CudaRng);
+unsafe impl Send for CudaRng {}
+
+#[derive(Clone)]
+pub struct CudaDevice {
+    id: DeviceId,
+    device: Arc<cudarc::driver::CudaDevice>,
+    pub(crate) blas: Arc<cudarc::cublas::CudaBlas>,
+    curand: Arc<Mutex<CudaRng>>,
+}
+
+impl std::fmt::Debug for CudaDevice {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "CudaDevice({:?})", self.id)
+    }
+}
+
+impl std::ops::Deref for CudaDevice {
+    type Target = Arc<cudarc::driver::CudaDevice>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.device
+    }
+}
+
+impl CudaDevice {
+    pub fn cuda_device(&self) -> Arc<cudarc::driver::CudaDevice> {
+        self.device.clone()
+    }
+
+    pub fn id(&self) -> DeviceId {
+        self.id
+    }
+
+    fn const_impl(&self, v: f64, shape: &Shape, dtype: DType) -> Result<CudaStorage> {
+        let elem_count = shape.elem_count();
+        let cfg = LaunchConfig::for_num_elems(elem_count as u32);
+        let slice = match dtype {
+            DType::U8 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<u8>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_u8", kernels::FILL)?;
+                let params = (&data, v as u8, elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::U8(data)
+            }
+            DType::U32 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<u32>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_u32", kernels::FILL)?;
+                let params = (&data, v as u32, elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::U32(data)
+            }
+            DType::I64 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<i64>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_i64", kernels::FILL)?;
+                let params = (&data, v as i64, elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::I64(data)
+            }
+            DType::BF16 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<bf16>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_bf16", kernels::FILL)?;
+                let params = (&data, bf16::from_f64(v), elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::BF16(data)
+            }
+            DType::F16 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<f16>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_f16", kernels::FILL)?;
+                let params = (&data, f16::from_f64(v), elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::F16(data)
+            }
+            DType::F32 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<f32>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_f32", kernels::FILL)?;
+                let params = (&data, v as f32, elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::F32(data)
+            }
+            DType::F64 => {
+                // SAFETY: Set later by running the fill kernel.
+                let data = unsafe { self.alloc::<f64>(elem_count) }.w()?;
+                let func = self.get_or_load_func("fill_f64", kernels::FILL)?;
+                let params = (&data, v, elem_count);
+                unsafe { func.launch(cfg, params) }.w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    pub fn get_or_load_func(&self, module_name: &str, ptx: &'static str) -> Result<CudaFunction> {
+        if !self.has_func(module_name, module_name) {
+            // Leaking the string here is a bit sad but we need a &'static str and this is only
+            // done once per kernel name.
+            let static_module_name = Box::leak(module_name.to_string().into_boxed_str());
+            self.load_ptx(ptx.into(), module_name, &[static_module_name])
+                .map_err(|cuda| CudaError::Load {
+                    cuda,
+                    module_name: module_name.to_string(),
+                })
+                .w()?;
+        }
+        self.get_func(module_name, module_name)
+            // Clippy recommends this `ok_or` rather than `ok_or_else` so hopefully the compiler is
+            // able to only build the error value if needed.
+            .ok_or(CudaError::MissingKernel {
+                module_name: module_name.to_string(),
+            })
+            .w()
+    }
+}
+
+impl BackendDevice for CudaDevice {
+    type Storage = CudaStorage;
+
+    fn new(ordinal: usize) -> Result<Self> {
+        let device = cudarc::driver::CudaDevice::new(ordinal).w()?;
+        let blas = cudarc::cublas::CudaBlas::new(device.clone()).w()?;
+        let curand = cudarc::curand::CudaRng::new(299792458, device.clone()).w()?;
+        Ok(Self {
+            id: DeviceId::new(),
+            device,
+            blas: Arc::new(blas),
+            curand: Arc::new(Mutex::new(CudaRng(curand))),
+        })
+    }
+
+    fn set_seed(&self, seed: u64) -> Result<()> {
+        // We do not call set_seed but instead create a new curand object. This ensures that the
+        // state will be identical and the same random numbers will be generated.
+        let mut curand = self.curand.lock().unwrap();
+        curand.0 = cudarc::curand::CudaRng::new(seed, self.device.clone()).w()?;
+        Ok(())
+    }
+
+    fn location(&self) -> crate::DeviceLocation {
+        crate::DeviceLocation::Cuda {
+            gpu_id: self.device.ordinal(),
+        }
+    }
+
+    fn same_device(&self, rhs: &Self) -> bool {
+        self.id == rhs.id
+    }
+
+    fn zeros_impl(&self, shape: &Shape, dtype: DType) -> Result<CudaStorage> {
+        let elem_count = shape.elem_count();
+        let slice = match dtype {
+            DType::U8 => {
+                let data = self.alloc_zeros::<u8>(elem_count).w()?;
+                CudaStorageSlice::U8(data)
+            }
+            DType::U32 => {
+                let data = self.alloc_zeros::<u32>(elem_count).w()?;
+                CudaStorageSlice::U32(data)
+            }
+            DType::I64 => {
+                let data = self.alloc_zeros::<i64>(elem_count).w()?;
+                CudaStorageSlice::I64(data)
+            }
+            DType::BF16 => {
+                let data = self.alloc_zeros::<bf16>(elem_count).w()?;
+                CudaStorageSlice::BF16(data)
+            }
+            DType::F16 => {
+                let data = self.alloc_zeros::<f16>(elem_count).w()?;
+                CudaStorageSlice::F16(data)
+            }
+            DType::F32 => {
+                let data = self.alloc_zeros::<f32>(elem_count).w()?;
+                CudaStorageSlice::F32(data)
+            }
+            DType::F64 => {
+                let data = self.alloc_zeros::<f64>(elem_count).w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    fn rand_uniform(&self, shape: &Shape, dtype: DType, lo: f64, up: f64) -> Result<CudaStorage> {
+        let elem_count = shape.elem_count();
+        let curand = self.curand.lock().unwrap();
+        let slice = match dtype {
+            // TODO: Add support for F16 and BF16 though this is likely to require some upstream
+            // cudarc changes.
+            DType::U8 | DType::U32 | DType::I64 | DType::F16 | DType::BF16 => {
+                Err(CudaError::UnsupportedDtype {
+                    dtype,
+                    op: "rand_uniform",
+                })
+                .w()?
+            }
+            DType::F32 => {
+                let mut data = unsafe { self.alloc::<f32>(elem_count) }.w()?;
+                curand.0.fill_with_uniform(&mut data).w()?;
+                CudaStorageSlice::F32(data)
+            }
+            DType::F64 => {
+                let mut data = unsafe { self.alloc::<f64>(elem_count) }.w()?;
+                curand.0.fill_with_uniform(&mut data).w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        let slice = if lo == 0. && up == 1.0 {
+            slice
+        } else {
+            use super::utils::Map1;
+            let layout = Layout::contiguous(shape);
+            super::Affine(up - lo, lo).map(&slice, self, &layout)?
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    fn rand_normal(&self, shape: &Shape, dtype: DType, mean: f64, std: f64) -> Result<CudaStorage> {
+        // TODO: Add support for F16 and BF16 though this is likely to require some upstream
+        // cudarc changes.
+        let elem_count = shape.elem_count();
+        let curand = self.curand.lock().unwrap();
+        // curand can only generate an odd number of values.
+        // https://github.com/huggingface/candle/issues/734
+        let elem_count_round = if elem_count % 2 == 1 {
+            elem_count + 1
+        } else {
+            elem_count
+        };
+        let slice = match dtype {
+            DType::U8 | DType::U32 | DType::I64 | DType::F16 | DType::BF16 => {
+                Err(CudaError::UnsupportedDtype {
+                    dtype,
+                    op: "rand_normal",
+                })
+                .w()?
+            }
+            DType::F32 => {
+                let mut data = unsafe { self.alloc::<f32>(elem_count_round) }.w()?;
+                curand
+                    .0
+                    .fill_with_normal(&mut data, mean as f32, std as f32)
+                    .w()?;
+                CudaStorageSlice::F32(data)
+            }
+            DType::F64 => {
+                let mut data = unsafe { self.alloc::<f64>(elem_count_round) }.w()?;
+                curand.0.fill_with_normal(&mut data, mean, std).w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    fn ones_impl(&self, shape: &Shape, dtype: DType) -> Result<CudaStorage> {
+        self.const_impl(1., shape, dtype)
+    }
+
+    unsafe fn alloc_uninit(&self, shape: &Shape, dtype: DType) -> Result<Self::Storage> {
+        let elem_count = shape.elem_count();
+        let slice = match dtype {
+            DType::U8 => {
+                let data = self.alloc::<u8>(elem_count).w()?;
+                CudaStorageSlice::U8(data)
+            }
+            DType::U32 => {
+                let data = self.alloc::<u32>(elem_count).w()?;
+                CudaStorageSlice::U32(data)
+            }
+            DType::I64 => {
+                let data = self.alloc::<i64>(elem_count).w()?;
+                CudaStorageSlice::I64(data)
+            }
+            DType::BF16 => {
+                let data = self.alloc::<bf16>(elem_count).w()?;
+                CudaStorageSlice::BF16(data)
+            }
+            DType::F16 => {
+                let data = self.alloc::<f16>(elem_count).w()?;
+                CudaStorageSlice::F16(data)
+            }
+            DType::F32 => {
+                let data = self.alloc::<f32>(elem_count).w()?;
+                CudaStorageSlice::F32(data)
+            }
+            DType::F64 => {
+                let data = self.alloc::<f64>(elem_count).w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    fn storage_from_cpu_storage(&self, storage: &CpuStorage) -> Result<CudaStorage> {
+        let slice = match storage {
+            CpuStorage::U8(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::U8(data)
+            }
+            CpuStorage::U32(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::U32(data)
+            }
+            CpuStorage::I64(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::I64(data)
+            }
+            CpuStorage::BF16(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::BF16(data)
+            }
+            CpuStorage::F16(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::F16(data)
+            }
+            CpuStorage::F32(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::F32(data)
+            }
+            CpuStorage::F64(storage) => {
+                let data = self.htod_sync_copy(storage).w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    fn storage_from_cpu_storage_owned(&self, storage: CpuStorage) -> Result<CudaStorage> {
+        let slice = match storage {
+            CpuStorage::U8(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::U8(data)
+            }
+            CpuStorage::U32(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::U32(data)
+            }
+            CpuStorage::I64(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::I64(data)
+            }
+            CpuStorage::BF16(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::BF16(data)
+            }
+            CpuStorage::F16(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::F16(data)
+            }
+            CpuStorage::F32(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::F32(data)
+            }
+            CpuStorage::F64(storage) => {
+                let data = self.htod_copy(storage).w()?;
+                CudaStorageSlice::F64(data)
+            }
+        };
+        Ok(CudaStorage {
+            slice,
+            device: self.clone(),
+        })
+    }
+
+    fn synchronize(&self) -> Result<()> {
+        self.device.synchronize().map_err(crate::Error::wrap)?;
+        Ok(())
+    }
+}

candle-core/src/cuda_backend/error.rs

@@ -0,0 +1,62 @@
+use crate::{DType, Layout};
+
+/// cudarc related errors
+#[derive(thiserror::Error, Debug)]
+pub enum CudaError {
+    #[error(transparent)]
+    Cuda(#[from] cudarc::driver::DriverError),
+
+    #[error(transparent)]
+    Compiler(#[from] cudarc::nvrtc::CompileError),
+
+    #[error(transparent)]
+    Cublas(#[from] cudarc::cublas::result::CublasError),
+
+    #[error(transparent)]
+    Curand(#[from] cudarc::curand::result::CurandError),
+
+    #[error("missing kernel '{module_name}'")]
+    MissingKernel { module_name: String },
+
+    #[error("unsupported dtype {dtype:?} for {op}")]
+    UnsupportedDtype { dtype: DType, op: &'static str },
+
+    #[error("internal error '{0}'")]
+    InternalError(&'static str),
+
+    #[error("matmul is only supported for contiguous tensors lstride: {lhs_stride:?} rstride: {rhs_stride:?} mnk: {mnk:?}")]
+    MatMulNonContiguous {
+        lhs_stride: Layout,
+        rhs_stride: Layout,
+        mnk: (usize, usize, usize),
+    },
+
+    #[error("{msg}, expected: {expected:?}, got: {got:?}")]
+    UnexpectedDType {
+        msg: &'static str,
+        expected: DType,
+        got: DType,
+    },
+
+    #[error("{cuda} when loading {module_name}")]
+    Load {
+        cuda: cudarc::driver::DriverError,
+        module_name: String,
+    },
+}
+
+impl From<CudaError> for crate::Error {
+    fn from(val: CudaError) -> Self {
+        crate::Error::Cuda(Box::new(val)).bt()
+    }
+}
+
+pub trait WrapErr<O> {
+    fn w(self) -> std::result::Result<O, crate::Error>;
+}
+
+impl<O, E: Into<CudaError>> WrapErr<O> for std::result::Result<O, E> {
+    fn w(self) -> std::result::Result<O, crate::Error> {
+        self.map_err(|e| crate::Error::Cuda(Box::new(e.into())).bt())
+    }
+}

candle-core/src/cuda_backend/mod.rs

@@ -5,395 +5,41 @@ pub use candle_kernels as kernels;
 pub use cudarc;
 use cudarc::cublas::{Gemm, GemmConfig, StridedBatchedConfig};
 use cudarc::driver::{
-    CudaFunction, CudaSlice, DevicePtr, DeviceRepr, DeviceSlice, LaunchAsync, LaunchConfig,
-    ValidAsZeroBits,
+    CudaSlice, DevicePtr, DeviceRepr, DeviceSlice, LaunchAsync, LaunchConfig, ValidAsZeroBits,
 };
 use half::{bf16, f16};
-use std::sync::{Arc, Mutex};
 
-/// cudarc related errors
-#[derive(thiserror::Error, Debug)]
-pub enum CudaError {
-    #[error(transparent)]
-    Cuda(#[from] cudarc::driver::DriverError),
+#[cfg(feature = "cudnn")]
+pub mod cudnn;
+mod device;
+mod error;
+mod utils;
+pub use device::{CudaDevice, DeviceId};
+pub use error::{CudaError, WrapErr};
+pub use utils::{Map1, Map1Any, Map2, Map2Any, Map2InPlace, S};
 
-    #[error(transparent)]
-    Compiler(#[from] cudarc::nvrtc::CompileError),
-
-    #[error(transparent)]
-    Cublas(#[from] cudarc::cublas::result::CublasError),
-
-    #[error(transparent)]
-    Curand(#[from] cudarc::curand::result::CurandError),
-
-    #[error("missing kernel '{module_name}'")]
-    MissingKernel { module_name: String },
-
-    #[error("unsupported dtype {dtype:?} for {op}")]
-    UnsupportedDtype { dtype: DType, op: &'static str },
-
-    #[error("internal error '{0}'")]
-    InternalError(&'static str),
-
-    #[error("matmul is only supported for contiguous tensors lstride: {lhs_stride:?} rstride: {rhs_stride:?} mnk: {mnk:?}")]
-    MatMulNonContiguous {
-        lhs_stride: Vec<usize>,
-        rhs_stride: Vec<usize>,
-        mnk: (usize, usize, usize),
-    },
-
-    #[error("{msg}, expected: {expected:?}, got: {got:?}")]
-    UnexpectedDType {
-        msg: &'static str,
-        expected: DType,
-        got: DType,
-    },
-
-    #[error("{cuda} when loading {module_name}")]
-    Load {
-        cuda: cudarc::driver::DriverError,
-        module_name: String,
-    },
+enum SlicePtrOrNull<T> {
+    Ptr(CudaSlice<T>),
+    Null,
 }
 
-impl From<CudaError> for crate::Error {
-    fn from(val: CudaError) -> Self {
-        crate::Error::Cuda(Box::new(val)).bt()
-    }
-}
-
-/// Unique identifier for cuda devices.
-#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
-pub struct DeviceId(usize);
-
-impl DeviceId {
-    fn new() -> Self {
-        // https://users.rust-lang.org/t/idiomatic-rust-way-to-generate-unique-id/33805
-        use std::sync::atomic;
-        static COUNTER: atomic::AtomicUsize = atomic::AtomicUsize::new(1);
-        Self(COUNTER.fetch_add(1, atomic::Ordering::Relaxed))
-    }
-}
-
-struct CudaRng(cudarc::curand::CudaRng);
-unsafe impl Send for CudaRng {}
-
-#[derive(Clone)]
-pub struct CudaDevice {
-    id: DeviceId,
-    device: Arc<cudarc::driver::CudaDevice>,
-    blas: Arc<cudarc::cublas::CudaBlas>,
-    curand: Arc<Mutex<CudaRng>>,
-}
-
-impl std::fmt::Debug for CudaDevice {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "CudaDevice({:?})", self.id)
-    }
-}
-
-impl std::ops::Deref for CudaDevice {
-    type Target = Arc<cudarc::driver::CudaDevice>;
-
-    fn deref(&self) -> &Self::Target {
-        &self.device
-    }
-}
-
-pub trait WrapErr<O> {
-    fn w(self) -> std::result::Result<O, crate::Error>;
-}
-
-impl<O, E: Into<CudaError>> WrapErr<O> for std::result::Result<O, E> {
-    fn w(self) -> std::result::Result<O, crate::Error> {
-        self.map_err(|e| crate::Error::Cuda(Box::new(e.into())))
-    }
-}
-
-impl CudaDevice {
-    pub fn cuda_device(&self) -> Arc<cudarc::driver::CudaDevice> {
-        self.device.clone()
-    }
-
-    pub fn id(&self) -> DeviceId {
-        self.id
-    }
-
-    fn const_impl(&self, v: f64, shape: &Shape, dtype: DType) -> Result<CudaStorage> {
-        let elem_count = shape.elem_count();
-        let cfg = LaunchConfig::for_num_elems(elem_count as u32);
-        let slice = match dtype {
-            DType::U8 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<u8>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_u8", kernels::FILL)?;
-                let params = (&data, v as u8, elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::U8(data)
-            }
-            DType::U32 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<u32>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_u32", kernels::FILL)?;
-                let params = (&data, v as u32, elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::U32(data)
-            }
-            DType::I64 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<i64>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_i64", kernels::FILL)?;
-                let params = (&data, v as i64, elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::I64(data)
-            }
-            DType::BF16 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<bf16>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_bf16", kernels::FILL)?;
-                let params = (&data, bf16::from_f64(v), elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::BF16(data)
-            }
-            DType::F16 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<f16>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_f16", kernels::FILL)?;
-                let params = (&data, f16::from_f64(v), elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::F16(data)
-            }
-            DType::F32 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<f32>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_f32", kernels::FILL)?;
-                let params = (&data, v as f32, elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::F32(data)
-            }
-            DType::F64 => {
-                // SAFETY: Set later by running the fill kernel.
-                let data = unsafe { self.alloc::<f64>(elem_count) }.w()?;
-                let func = self.get_or_load_func("fill_f64", kernels::FILL)?;
-                let params = (&data, v, elem_count);
-                unsafe { func.launch(cfg, params) }.w()?;
-                CudaStorageSlice::F64(data)
-            }
-        };
-        Ok(CudaStorage {
-            slice,
-            device: self.clone(),
-        })
-    }
-
-    pub fn get_or_load_func(&self, module_name: &str, ptx: &'static str) -> Result<CudaFunction> {
-        if !self.has_func(module_name, module_name) {
-            // Leaking the string here is a bit sad but we need a &'static str and this is only
-            // done once per kernel name.
-            let static_module_name = Box::leak(module_name.to_string().into_boxed_str());
-            self.load_ptx(ptx.into(), module_name, &[static_module_name])
-                .map_err(|cuda| CudaError::Load {
-                    cuda,
-                    module_name: module_name.to_string(),
-                })
-                .w()?;
-        }
-        self.get_func(module_name, module_name)
-            // Clippy recommends this `ok_or` rather than `ok_or_else` so hopefully the compiler is
-            // able to only build the error value if needed.
-            .ok_or(CudaError::MissingKernel {
-                module_name: module_name.to_string(),
-            })
-            .w()
-    }
-}
-
-impl BackendDevice for CudaDevice {
-    type Storage = CudaStorage;
-
-    fn new(ordinal: usize) -> Result<Self> {
-        let device = cudarc::driver::CudaDevice::new(ordinal).w()?;
-        let blas = cudarc::cublas::CudaBlas::new(device.clone()).w()?;
-        let curand = cudarc::curand::CudaRng::new(299792458, device.clone()).w()?;
-        Ok(Self {
-            id: DeviceId::new(),
-            device,
-            blas: Arc::new(blas),
-            curand: Arc::new(Mutex::new(CudaRng(curand))),
-        })
-    }
-
-    fn set_seed(&self, seed: u64) -> Result<()> {
-        // We do not call set_seed but instead create a new curand object. This ensures that the
-        // state will be identical and the same random numbers will be generated.
-        let mut curand = self.curand.lock().unwrap();
-        curand.0 = cudarc::curand::CudaRng::new(seed, self.device.clone()).w()?;
-        Ok(())
-    }
-
-    fn location(&self) -> crate::DeviceLocation {
-        crate::DeviceLocation::Cuda {
-            gpu_id: self.device.ordinal(),
+unsafe impl<T: DeviceRepr> DeviceRepr for &SlicePtrOrNull<T> {
+    fn as_kernel_param(&self) -> *mut std::ffi::c_void {
+        match self {
+            SlicePtrOrNull::Ptr(slice) => slice.as_kernel_param(),
+            SlicePtrOrNull::Null => 0usize.as_kernel_param(),
         }
     }
+}
 
-    fn same_device(&self, rhs: &Self) -> bool {
-        self.id == rhs.id
-    }
-
-    fn zeros_impl(&self, shape: &Shape, dtype: DType) -> Result<CudaStorage> {
-        let elem_count = shape.elem_count();
-        let slice = match dtype {
-            DType::U8 => {
-                let data = self.alloc_zeros::<u8>(elem_count).w()?;
-                CudaStorageSlice::U8(data)
-            }
-            DType::U32 => {
-                let data = self.alloc_zeros::<u32>(elem_count).w()?;
-                CudaStorageSlice::U32(data)
-            }
-            DType::I64 => {
-                let data = self.alloc_zeros::<i64>(elem_count).w()?;
-                CudaStorageSlice::I64(data)
-            }
-            DType::BF16 => {
-                let data = self.alloc_zeros::<bf16>(elem_count).w()?;
-                CudaStorageSlice::BF16(data)
-            }
-            DType::F16 => {
-                let data = self.alloc_zeros::<f16>(elem_count).w()?;
-                CudaStorageSlice::F16(data)
-            }
-            DType::F32 => {
-                let data = self.alloc_zeros::<f32>(elem_count).w()?;
-                CudaStorageSlice::F32(data)
-            }
-            DType::F64 => {
-                let data = self.alloc_zeros::<f64>(elem_count).w()?;
-                CudaStorageSlice::F64(data)
-            }
-        };
-        Ok(CudaStorage {
-            slice,
-            device: self.clone(),
-        })
-    }
-
-    fn rand_uniform(&self, shape: &Shape, dtype: DType, lo: f64, up: f64) -> Result<CudaStorage> {
-        let elem_count = shape.elem_count();
-        let curand = self.curand.lock().unwrap();
-        let slice = match dtype {
-            // TODO: Add support for F16 and BF16 though this is likely to require some upstream
-            // cudarc changes.
-            DType::U8 | DType::U32 | DType::I64 | DType::F16 | DType::BF16 => {
-                Err(CudaError::UnsupportedDtype {
-                    dtype,
-                    op: "rand_uniform",
-                })
-                .w()?
-            }
-            DType::F32 => {
-                let mut data = unsafe { self.alloc::<f32>(elem_count) }.w()?;
-                curand.0.fill_with_uniform(&mut data).w()?;
-                CudaStorageSlice::F32(data)
-            }
-            DType::F64 => {
-                let mut data = unsafe { self.alloc::<f64>(elem_count) }.w()?;
-                curand.0.fill_with_uniform(&mut data).w()?;
-                CudaStorageSlice::F64(data)
-            }
-        };
-        let slice = if lo == 0. && up == 1.0 {
-            slice
+impl SlicePtrOrNull<usize> {
+    fn params_from_layout(dev: &CudaDevice, l: &Layout) -> Result<Self> {
+        let ds = if l.is_contiguous() {
+            SlicePtrOrNull::Null
         } else {
-            let layout = Layout::contiguous(shape);
-            Affine(up - lo, lo).map(&slice, self, &layout)?
+            SlicePtrOrNull::Ptr(dev.htod_copy([l.dims(), l.stride()].concat()).w()?)
         };
-        Ok(CudaStorage {
-            slice,
-            device: self.clone(),
-        })
-    }
-
-    fn rand_normal(&self, shape: &Shape, dtype: DType, mean: f64, std: f64) -> Result<CudaStorage> {
-        // TODO: Add support for F16 and BF16 though this is likely to require some upstream
-        // cudarc changes.
-        let elem_count = shape.elem_count();
-        let curand = self.curand.lock().unwrap();
-        // curand can only generate an odd number of values.
-        // https://github.com/huggingface/candle/issues/734
-        let elem_count_round = if elem_count % 2 == 1 {
-            elem_count + 1
-        } else {
-            elem_count
-        };
-        let slice = match dtype {
-            DType::U8 | DType::U32 | DType::I64 | DType::F16 | DType::BF16 => {
-                Err(CudaError::UnsupportedDtype {
-                    dtype,
-                    op: "rand_normal",
-                })
-                .w()?
-            }
-            DType::F32 => {
-                let mut data = unsafe { self.alloc::<f32>(elem_count_round) }.w()?;
-                curand
-                    .0
-                    .fill_with_normal(&mut data, mean as f32, std as f32)
-                    .w()?;
-                CudaStorageSlice::F32(data)
-            }
-            DType::F64 => {
-                let mut data = unsafe { self.alloc::<f64>(elem_count_round) }.w()?;
-                curand.0.fill_with_normal(&mut data, mean, std).w()?;
-                CudaStorageSlice::F64(data)
-            }
-        };
-        Ok(CudaStorage {
-            slice,
-            device: self.clone(),
-        })
-    }
-
-    fn ones_impl(&self, shape: &Shape, dtype: DType) -> Result<CudaStorage> {
-        self.const_impl(1., shape, dtype)
-    }
-
-    fn storage_from_cpu_storage(&self, storage: &CpuStorage) -> Result<CudaStorage> {
-        let slice = match storage {
-            CpuStorage::U8(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::U8(data)
-            }
-            CpuStorage::U32(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::U32(data)
-            }
-            CpuStorage::I64(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::I64(data)
-            }
-            CpuStorage::BF16(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::BF16(data)
-            }
-            CpuStorage::F16(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::F16(data)
-            }
-            CpuStorage::F32(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::F32(data)
-            }
-            CpuStorage::F64(storage) => {
-                let data = self.htod_sync_copy(storage).w()?;
-                CudaStorageSlice::F64(data)
-            }
-        };
-        Ok(CudaStorage {
-            slice,
-            device: self.clone(),
-        })
+        Ok(ds)
     }
 }
 
@@ -407,133 +53,6 @@ pub enum CudaStorageSlice {
     F32(CudaSlice<f32>),
     F64(CudaSlice<f64>),
 }
-type S = CudaStorageSlice;
-
-pub trait Map1 {
-    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
-        &self,
-        src: &CudaSlice<T>,
-        dev: &CudaDevice,
-        layout: &Layout,
-    ) -> Result<CudaSlice<T>>;
-
-    fn map(&self, s: &S, d: &CudaDevice, l: &Layout) -> Result<S> {
-        let out = match s {
-            S::U8(s) => S::U8(self.f(s, d, l)?),
-            S::U32(s) => S::U32(self.f(s, d, l)?),
-            S::I64(s) => S::I64(self.f(s, d, l)?),
-            S::BF16(s) => S::BF16(self.f(s, d, l)?),
-            S::F16(s) => S::F16(self.f(s, d, l)?),
-            S::F32(s) => S::F32(self.f(s, d, l)?),
-            S::F64(s) => S::F64(self.f(s, d, l)?),
-        };
-        Ok(out)
-    }
-}
-
-pub trait Map2 {
-    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
-        &self,
-        src1: &CudaSlice<T>,
-        layout1: &Layout,
-        src2: &CudaSlice<T>,
-        layout2: &Layout,
-        dev: &CudaDevice,
-    ) -> Result<CudaSlice<T>>;
-
-    fn map(&self, s1: &S, l1: &Layout, s2: &S, l2: &Layout, d: &CudaDevice) -> Result<S> {
-        let out = match (s1, s2) {
-            (S::U8(s1), S::U8(s2)) => S::U8(self.f(s1, l1, s2, l2, d)?),
-            (S::U32(s1), S::U32(s2)) => S::U32(self.f(s1, l1, s2, l2, d)?),
-            (S::I64(s1), S::I64(s2)) => S::I64(self.f(s1, l1, s2, l2, d)?),
-            (S::BF16(s1), S::BF16(s2)) => S::BF16(self.f(s1, l1, s2, l2, d)?),
-            (S::F16(s1), S::F16(s2)) => S::F16(self.f(s1, l1, s2, l2, d)?),
-            (S::F32(s1), S::F32(s2)) => S::F32(self.f(s1, l1, s2, l2, d)?),
-            (S::F64(s1), S::F64(s2)) => S::F64(self.f(s1, l1, s2, l2, d)?),
-            _ => Err(CudaError::InternalError("dtype mismatch in binary op"))?,
-        };
-        Ok(out)
-    }
-}
-
-pub trait Map2InPlace {
-    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
-        &self,
-        dst: &mut CudaSlice<T>,
-        dst_shape: &Shape,
-        src: &CudaSlice<T>,
-        src_l: &Layout,
-        dev: &CudaDevice,
-    ) -> Result<()>;
-
-    fn map(
-        &self,
-        dst: &mut S,
-        dst_s: &Shape,
-        src: &S,
-        src_l: &Layout,
-        d: &CudaDevice,
-    ) -> Result<()> {
-        match (dst, src) {
-            (S::U8(dst), S::U8(src)) => self.f(dst, dst_s, src, src_l, d),
-            (S::U32(dst), S::U32(src)) => self.f(dst, dst_s, src, src_l, d),
-            (S::I64(dst), S::I64(src)) => self.f(dst, dst_s, src, src_l, d),
-            (S::BF16(dst), S::BF16(src)) => self.f(dst, dst_s, src, src_l, d),
-            (S::F16(dst), S::F16(src)) => self.f(dst, dst_s, src, src_l, d),
-            (S::F32(dst), S::F32(src)) => self.f(dst, dst_s, src, src_l, d),
-            (S::F64(dst), S::F64(src)) => self.f(dst, dst_s, src, src_l, d),
-            _ => Err(CudaError::InternalError("dtype mismatch in binary op"))?,
-        }
-    }
-}
-
-pub trait Map1Any {
-    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits, W: Fn(CudaSlice<T>) -> S>(
-        &self,
-        src: &CudaSlice<T>,
-        dev: &CudaDevice,
-        layout: &Layout,
-        wrap: W,
-    ) -> Result<S>;
-
-    fn map(&self, s: &S, d: &CudaDevice, l: &Layout) -> Result<S> {
-        let out = match s {
-            S::U8(s) => self.f(s, d, l, S::U8)?,
-            S::U32(s) => self.f(s, d, l, S::U32)?,
-            S::I64(s) => self.f(s, d, l, S::I64)?,
-            S::BF16(s) => self.f(s, d, l, S::BF16)?,
-            S::F16(s) => self.f(s, d, l, S::F16)?,
-            S::F32(s) => self.f(s, d, l, S::F32)?,
-            S::F64(s) => self.f(s, d, l, S::F64)?,
-        };
-        Ok(out)
-    }
-}
-
-pub trait Map2Any {
-    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
-        &self,
-        src1: &CudaSlice<T>,
-        layout1: &Layout,
-        src2: &CudaSlice<T>,
-        layout2: &Layout,
-        dev: &CudaDevice,
-    ) -> Result<S>;
-
-    fn map(&self, s1: &S, l1: &Layout, s2: &S, l2: &Layout, d: &CudaDevice) -> Result<S> {
-        let out = match (s1, s2) {
-            (S::U8(s1), S::U8(s2)) => self.f(s1, l1, s2, l2, d)?,
-            (S::U32(s1), S::U32(s2)) => self.f(s1, l1, s2, l2, d)?,
-            (S::I64(s1), S::I64(s2)) => self.f(s1, l1, s2, l2, d)?,
-            (S::BF16(s1), S::BF16(s2)) => self.f(s1, l1, s2, l2, d)?,
-            (S::F16(s1), S::F16(s2)) => self.f(s1, l1, s2, l2, d)?,
-            (S::F32(s1), S::F32(s2)) => self.f(s1, l1, s2, l2, d)?,
-            (S::F64(s1), S::F64(s2)) => self.f(s1, l1, s2, l2, d)?,
-            _ => Err(CudaError::InternalError("dtype mismatch in binary op")).w()?,
-        };
-        Ok(out)
-    }
-}
 
 struct Clone;
 impl Map1 for Clone {
@@ -564,7 +83,7 @@ impl Map1 for Affine {
         let dims = shape.dims();
         let el = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(el as u32);
-        let ds = dev.htod_copy([dims, layout.stride()].concat()).w()?;
+        let ds = SlicePtrOrNull::params_from_layout(dev, layout)?;
         let src = &src.slice(layout.start_offset()..);
         let func = dev.get_or_load_func(&kernel_name::<T>("affine"), kernels::AFFINE)?;
         // SAFETY: Set later by running the kernel.
@@ -596,7 +115,7 @@ impl Map1 for Elu {
         let dims = shape.dims();
         let el = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(el as u32);
-        let ds = dev.htod_copy([dims, layout.stride()].concat()).w()?;
+        let ds = SlicePtrOrNull::params_from_layout(dev, layout)?;
         let src = &src.slice(layout.start_offset()..);
         let func = dev.get_or_load_func(&kernel_name::<T>("uelu"), kernels::UNARY)?;
         // SAFETY: Set later by running the kernel.
@@ -719,7 +238,7 @@ impl Map1 for Powf {
         let dims = shape.dims();
         let el = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(el as u32);
-        let ds = dev.htod_copy([dims, layout.stride()].concat()).w()?;
+        let ds = SlicePtrOrNull::params_from_layout(dev, layout)?;
         let src = &src.slice(layout.start_offset()..);
         let func = dev.get_or_load_func(&kernel_name::<T>("upowf"), kernels::UNARY)?;
         // SAFETY: Set later by running the kernel.
@@ -852,7 +371,7 @@ impl<U: UnaryOpT> Map1 for U {
         let dims = shape.dims();
         let el_count = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(el_count as u32);
-        let ds = dev.htod_copy([dims, layout.stride()].concat()).w()?;
+        let ds = SlicePtrOrNull::params_from_layout(dev, layout)?;
         let src = &src.slice(layout.start_offset()..);
         let func = dev.get_or_load_func(&kernel_name::<T>(U::KERNEL), kernels::UNARY)?;
         // SAFETY: Set later by running the kernel.
@@ -1402,9 +921,14 @@ impl<U: crate::op::BinaryOpT> Map2 for U {
         let dims = shape.dims();
         let elem_count = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(elem_count as u32);
-        let dims_and_strides = dev
-            .htod_copy([dims, lhs_l.stride(), rhs_l.stride()].concat())
-            .w()?;
+        let dims_and_strides = if lhs_l.is_contiguous() && rhs_l.is_contiguous() {
+            SlicePtrOrNull::Null
+        } else {
+            SlicePtrOrNull::Ptr(
+                dev.htod_copy([dims, lhs_l.stride(), rhs_l.stride()].concat())
+                    .w()?,
+            )
+        };
         let lhs = &lhs.slice(lhs_l.start_offset()..);
         let rhs = &rhs.slice(rhs_l.start_offset()..);
         let func = dev.get_or_load_func(&kernel_name::<T>(U::KERNEL), kernels::BINARY)?;
@@ -1431,9 +955,14 @@ impl Map2Any for Cmp {
         let dims = shape.dims();
         let elem_count = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(elem_count as u32);
-        let dims_and_strides = dev
-            .htod_copy([dims, lhs_l.stride(), rhs_l.stride()].concat())
-            .w()?;
+        let dims_and_strides = if lhs_l.is_contiguous() && rhs_l.is_contiguous() {
+            SlicePtrOrNull::Null
+        } else {
+            SlicePtrOrNull::Ptr(
+                dev.htod_copy([dims, lhs_l.stride(), rhs_l.stride()].concat())
+                    .w()?,
+            )
+        };
         let lhs = &lhs.slice(lhs_l.start_offset()..);
         let rhs = &rhs.slice(rhs_l.start_offset()..);
         let name = match self.0 {
@@ -1541,26 +1070,30 @@ fn gemm_config<T>(
     let lhs_m1 = lhs_stride[lhs_stride.len() - 1];
     let lhs_m2 = lhs_stride[lhs_stride.len() - 2];
     // The a tensor has dims batching, k, n (rhs)
-    let (lda, transa) = if rhs_m1 == 1 && rhs_m2 == n {
+    // We also allow for the case where the stride on the minor dimension is not as expected but
+    // there is a single element.
+    let (lda, transa) = if (rhs_m1 == 1 || n == 1) && (rhs_m2 == n || k == 1) {
         (n as i32, cublasOperation_t::CUBLAS_OP_N)
-    } else if rhs_m1 == k && rhs_m2 == 1 {
+    } else if (rhs_m1 == k || n == 1) && (rhs_m2 == 1 || k == 1) {
         (k as i32, cublasOperation_t::CUBLAS_OP_T)
     } else {
         Err(CudaError::MatMulNonContiguous {
-            lhs_stride: lhs_stride.to_vec(),
-            rhs_stride: rhs_stride.to_vec(),
+            lhs_stride: lhs_l.clone(),
+            rhs_stride: rhs_l.clone(),
             mnk: (m, n, k),
         })?
     };
     // The b tensor has dims batching, m, k (lhs)
-    let (ldb, transb) = if lhs_m1 == 1 && lhs_m2 == k {
+    // We also allow for the case where the stride on the minor dimension is not as expected but
+    // there is a single element.
+    let (ldb, transb) = if (lhs_m1 == 1 || k == 1) && (lhs_m2 == k || m == 1) {
         (k as i32, cublasOperation_t::CUBLAS_OP_N)
-    } else if lhs_m1 == m && lhs_m2 == 1 {
+    } else if (lhs_m1 == m || k == 1) && (lhs_m2 == 1 || m == 1) {
         (m as i32, cublasOperation_t::CUBLAS_OP_T)
     } else {
         Err(CudaError::MatMulNonContiguous {
-            lhs_stride: lhs_stride.to_vec(),
-            rhs_stride: rhs_stride.to_vec(),
+            lhs_stride: lhs_l.clone(),
+            rhs_stride: rhs_l.clone(),
             mnk: (m, n, k),
         })?
     };
@@ -1581,21 +1114,25 @@ fn gemm_config<T>(
 
     let stride_b: usize = match lhs_stride[..lhs_stride.len() - 2] {
         [s1, stride] if s1 == stride * lhs_l.dims()[1] => stride,
+        [_, stride] if lhs_l.dims()[0] == 1 => stride,
+        [stride, _] if lhs_l.dims()[1] == 1 => stride,
         [stride] => stride,
         [] => m * k,
         _ => Err(CudaError::MatMulNonContiguous {
-            lhs_stride: lhs_stride.to_vec(),
-            rhs_stride: rhs_stride.to_vec(),
+            lhs_stride: lhs_l.clone(),
+            rhs_stride: rhs_l.clone(),
             mnk: (m, n, k),
         })?,
     };
     let stride_a: usize = match rhs_stride[..rhs_stride.len() - 2] {
         [s1, stride] if s1 == stride * rhs_l.dims()[1] => stride,
+        [_, stride] if rhs_l.dims()[0] == 1 => stride,
+        [stride, _] if rhs_l.dims()[1] == 1 => stride,
         [stride] => stride,
         [] => n * k,
         _ => Err(CudaError::MatMulNonContiguous {
-            lhs_stride: lhs_stride.to_vec(),
-            rhs_stride: rhs_stride.to_vec(),
+            lhs_stride: lhs_l.clone(),
+            rhs_stride: rhs_l.clone(),
             mnk: (m, n, k),
         })?,
     };
@@ -1640,7 +1177,7 @@ impl BackendStorage for CudaStorage {
         let el = shape.elem_count();
         let cfg = LaunchConfig::for_num_elems(el as u32);
         let dev = self.device();
-        let ds = dev.htod_copy([dims, layout.stride()].concat()).w()?;
+        let ds = SlicePtrOrNull::params_from_layout(dev, layout)?;
         let start_o = layout.start_offset();
         // This returns an i64 rather than a &i64, this is useful to get around some temporary
         // lifetime issue and is safe as long as self.slice does not go out of scope before inp
@@ -1844,7 +1381,10 @@ impl BackendStorage for CudaStorage {
             col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
         } else {
             // Make the kernel contiguous if not already the case.
-            let mut kernel_c = self.device().zeros_impl(kernel_l.shape(), kernel.dtype())?;
+            let mut kernel_c = unsafe {
+                self.device()
+                    .alloc_uninit(kernel_l.shape(), kernel.dtype())?
+            };
             kernel.copy_strided_src(&mut kernel_c, 0, kernel_l)?;
             let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
                 .transpose(1, 2)?
@@ -1852,7 +1392,7 @@ impl BackendStorage for CudaStorage {
             col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
         };
         let res_l = Layout::contiguous((b, l_out, n)).transpose(1, 2)?;
-        let mut res_t = self.device().zeros_impl(res_l.shape(), res.dtype())?;
+        let mut res_t = unsafe { self.device().alloc_uninit(res_l.shape(), res.dtype())? };
         res.copy_strided_src(&mut res_t, 0, &res_l)?;
         Ok(res_t)
     }
@@ -1909,7 +1449,10 @@ impl BackendStorage for CudaStorage {
             col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
         } else {
             // Make the kernel contiguous if not already the case.
-            let mut kernel_c = self.device().zeros_impl(kernel_l.shape(), kernel.dtype())?;
+            let mut kernel_c = unsafe {
+                self.device()
+                    .alloc_uninit(kernel_l.shape(), kernel.dtype())?
+            };
             kernel.copy_strided_src(&mut kernel_c, 0, kernel_l)?;
             let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
                 .transpose(1, 2)?
@@ -1919,7 +1462,7 @@ impl BackendStorage for CudaStorage {
         let res_l = Layout::contiguous((b, h_out, w_out, n))
             .transpose(1, 2)?
             .transpose(1, 3)?;
-        let mut res_t = self.device().zeros_impl(res_l.shape(), res.dtype())?;
+        let mut res_t = unsafe { self.device().alloc_uninit(res_l.shape(), res.dtype())? };
         res.copy_strided_src(&mut res_t, 0, &res_l)?;
         Ok(res_t)
     }
@@ -2056,7 +1599,7 @@ impl BackendStorage for CudaStorage {
         dim: usize,
     ) -> Result<Self> {
         let device = self.device().clone();
-        let mut acc = device.zeros_impl(l.shape(), self.dtype())?;
+        let mut acc = unsafe { device.alloc_uninit(l.shape(), self.dtype())? };
         self.copy_strided_src(&mut acc, 0, l)?;
         ScatterAdd(ids, ids_l, dim).map(&mut acc.slice, l.shape(), &src.slice, src_l, &device)?;
         Ok(acc)
@@ -2071,7 +1614,7 @@ impl BackendStorage for CudaStorage {
         dim: usize,
     ) -> Result<Self> {
         let device = self.device().clone();
-        let mut acc = device.zeros_impl(l.shape(), self.dtype())?;
+        let mut acc = unsafe { device.alloc_uninit(l.shape(), self.dtype())? };
         self.copy_strided_src(&mut acc, 0, l)?;
         IndexAdd(ids, ids_l, dim).map(&mut acc.slice, l.shape(), &src.slice, src_l, &device)?;
         Ok(acc)
@@ -2158,6 +1701,11 @@ impl BackendStorage for CudaStorage {
         let dev = &self.device;
         let d1 = d1 as u32;
         let d2 = d2 as u32;
+        // Nothing to copy so we exit early to avoid launching a kernel and some potential invalid
+        // argument with a null pointer.
+        if d1 == 0 || d2 == 0 {
+            return Ok(());
+        }
         let dst_s = dst_s as u32;
         let src_s = src_s as u32;
         let (src, dst, kname) = match (&self.slice, &mut dst.slice) {
@@ -2215,7 +1763,7 @@ impl BackendStorage for CudaStorage {
         }
         let cfg = LaunchConfig::for_num_elems(el_count as u32);
         let dev = &self.device;
-        let ds = dev.htod_copy([dims, src_l.stride()].concat()).w()?;
+        let ds = SlicePtrOrNull::params_from_layout(dev, src_l)?;
         match (&self.slice, &mut dst.slice) {
             (CudaStorageSlice::BF16(src), CudaStorageSlice::BF16(dst)) => {
                 let (src, mut dst) = slice_src_and_dst(src, src_l, dst, dst_offset);

candle-core/src/cuda_backend/utils.rs

@@ -0,0 +1,134 @@
+/// Helper functions to plug cuda kernels in candle.
+use crate::{Layout, Result, Shape, WithDType};
+pub use cudarc;
+use cudarc::driver::{CudaSlice, DeviceRepr, ValidAsZeroBits};
+
+use super::{CudaDevice, CudaError, WrapErr};
+
+pub type S = super::CudaStorageSlice;
+
+pub trait Map1 {
+    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
+        &self,
+        src: &CudaSlice<T>,
+        dev: &CudaDevice,
+        layout: &Layout,
+    ) -> Result<CudaSlice<T>>;
+
+    fn map(&self, s: &S, d: &CudaDevice, l: &Layout) -> Result<S> {
+        let out = match s {
+            S::U8(s) => S::U8(self.f(s, d, l)?),
+            S::U32(s) => S::U32(self.f(s, d, l)?),
+            S::I64(s) => S::I64(self.f(s, d, l)?),
+            S::BF16(s) => S::BF16(self.f(s, d, l)?),
+            S::F16(s) => S::F16(self.f(s, d, l)?),
+            S::F32(s) => S::F32(self.f(s, d, l)?),
+            S::F64(s) => S::F64(self.f(s, d, l)?),
+        };
+        Ok(out)
+    }
+}
+
+pub trait Map2 {
+    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
+        &self,
+        src1: &CudaSlice<T>,
+        layout1: &Layout,
+        src2: &CudaSlice<T>,
+        layout2: &Layout,
+        dev: &CudaDevice,
+    ) -> Result<CudaSlice<T>>;
+
+    fn map(&self, s1: &S, l1: &Layout, s2: &S, l2: &Layout, d: &CudaDevice) -> Result<S> {
+        let out = match (s1, s2) {
+            (S::U8(s1), S::U8(s2)) => S::U8(self.f(s1, l1, s2, l2, d)?),
+            (S::U32(s1), S::U32(s2)) => S::U32(self.f(s1, l1, s2, l2, d)?),
+            (S::I64(s1), S::I64(s2)) => S::I64(self.f(s1, l1, s2, l2, d)?),
+            (S::BF16(s1), S::BF16(s2)) => S::BF16(self.f(s1, l1, s2, l2, d)?),
+            (S::F16(s1), S::F16(s2)) => S::F16(self.f(s1, l1, s2, l2, d)?),
+            (S::F32(s1), S::F32(s2)) => S::F32(self.f(s1, l1, s2, l2, d)?),
+            (S::F64(s1), S::F64(s2)) => S::F64(self.f(s1, l1, s2, l2, d)?),
+            _ => Err(CudaError::InternalError("dtype mismatch in binary op"))?,
+        };
+        Ok(out)
+    }
+}
+
+pub trait Map2InPlace {
+    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
+        &self,
+        dst: &mut CudaSlice<T>,
+        dst_shape: &Shape,
+        src: &CudaSlice<T>,
+        src_l: &Layout,
+        dev: &CudaDevice,
+    ) -> Result<()>;
+
+    fn map(
+        &self,
+        dst: &mut S,
+        dst_s: &Shape,
+        src: &S,
+        src_l: &Layout,
+        d: &CudaDevice,
+    ) -> Result<()> {
+        match (dst, src) {
+            (S::U8(dst), S::U8(src)) => self.f(dst, dst_s, src, src_l, d),
+            (S::U32(dst), S::U32(src)) => self.f(dst, dst_s, src, src_l, d),
+            (S::I64(dst), S::I64(src)) => self.f(dst, dst_s, src, src_l, d),
+            (S::BF16(dst), S::BF16(src)) => self.f(dst, dst_s, src, src_l, d),
+            (S::F16(dst), S::F16(src)) => self.f(dst, dst_s, src, src_l, d),
+            (S::F32(dst), S::F32(src)) => self.f(dst, dst_s, src, src_l, d),
+            (S::F64(dst), S::F64(src)) => self.f(dst, dst_s, src, src_l, d),
+            _ => Err(CudaError::InternalError("dtype mismatch in binary op"))?,
+        }
+    }
+}
+
+pub trait Map1Any {
+    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits, W: Fn(CudaSlice<T>) -> S>(
+        &self,
+        src: &CudaSlice<T>,
+        dev: &CudaDevice,
+        layout: &Layout,
+        wrap: W,
+    ) -> Result<S>;
+
+    fn map(&self, s: &S, d: &CudaDevice, l: &Layout) -> Result<S> {
+        let out = match s {
+            S::U8(s) => self.f(s, d, l, S::U8)?,
+            S::U32(s) => self.f(s, d, l, S::U32)?,
+            S::I64(s) => self.f(s, d, l, S::I64)?,
+            S::BF16(s) => self.f(s, d, l, S::BF16)?,
+            S::F16(s) => self.f(s, d, l, S::F16)?,
+            S::F32(s) => self.f(s, d, l, S::F32)?,
+            S::F64(s) => self.f(s, d, l, S::F64)?,
+        };
+        Ok(out)
+    }
+}
+
+pub trait Map2Any {
+    fn f<T: DeviceRepr + WithDType + ValidAsZeroBits>(
+        &self,
+        src1: &CudaSlice<T>,
+        layout1: &Layout,
+        src2: &CudaSlice<T>,
+        layout2: &Layout,
+        dev: &CudaDevice,
+    ) -> Result<S>;
+
+    fn map(&self, s1: &S, l1: &Layout, s2: &S, l2: &Layout, d: &CudaDevice) -> Result<S> {
+        let out = match (s1, s2) {
+            (S::U8(s1), S::U8(s2)) => self.f(s1, l1, s2, l2, d)?,
+            (S::U32(s1), S::U32(s2)) => self.f(s1, l1, s2, l2, d)?,
+            (S::I64(s1), S::I64(s2)) => self.f(s1, l1, s2, l2, d)?,
+            (S::BF16(s1), S::BF16(s2)) => self.f(s1, l1, s2, l2, d)?,
+            (S::F16(s1), S::F16(s2)) => self.f(s1, l1, s2, l2, d)?,
+            (S::F32(s1), S::F32(s2)) => self.f(s1, l1, s2, l2, d)?,
+            (S::F64(s1), S::F64(s2)) => self.f(s1, l1, s2, l2, d)?,
+            _ => Err(CudaError::InternalError("dtype mismatch in binary op")).w()?,
+        };
+        Ok(out)
+    }
+}

candle-core/src/custom_op.rs

@@ -0,0 +1,377 @@
+use crate::op::{BackpropOp, Op};
+use crate::tensor::from_storage;
+use crate::{CpuStorage, CudaStorage, Layout, MetalStorage, Result, Shape, Tensor};
+use std::sync::Arc;
+
+/// Unary ops that can be defined in user-land.
+pub trait CustomOp1 {
+    // Box<dyn> does not support const yet, so use a function to get the name.
+    fn name(&self) -> &'static str;
+
+    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cpu_fwd(&self, storage: &CpuStorage, layout: &Layout) -> Result<(CpuStorage, Shape)>;
+
+    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cuda_fwd(&self, _storage: &CudaStorage, _layout: &Layout) -> Result<(CudaStorage, Shape)> {
+        Err(crate::Error::Cuda(
+            format!("no cuda implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn metal_fwd(
+        &self,
+        _storage: &MetalStorage,
+        _layout: &Layout,
+    ) -> Result<(MetalStorage, Shape)> {
+        Err(crate::Error::Metal(
+            format!("no metal implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// This function takes as argument the argument `arg` used in the forward pass, the result
+    /// produced by the forward operation `res` and the gradient of the result `grad_res`.
+    /// The function should return the gradient of the argument.
+    fn bwd(&self, _arg: &Tensor, _res: &Tensor, _grad_res: &Tensor) -> Result<Option<Tensor>> {
+        Err(crate::Error::BackwardNotSupported { op: self.name() })
+    }
+}
+
+pub trait CustomOp2 {
+    fn name(&self) -> &'static str;
+
+    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cpu_fwd(
+        &self,
+        s1: &CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+    ) -> Result<(CpuStorage, Shape)>;
+
+    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cuda_fwd(
+        &self,
+        _: &CudaStorage,
+        _: &Layout,
+        _: &CudaStorage,
+        _: &Layout,
+    ) -> Result<(CudaStorage, Shape)> {
+        Err(crate::Error::Cuda(
+            format!("no cuda implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn metal_fwd(
+        &self,
+        _: &MetalStorage,
+        _: &Layout,
+        _: &MetalStorage,
+        _: &Layout,
+    ) -> Result<(MetalStorage, Shape)> {
+        Err(crate::Error::Metal(
+            format!("no metal implementation for {}", self.name()).into(),
+        ))
+    }
+
+    fn bwd(
+        &self,
+        _arg1: &Tensor,
+        _arg2: &Tensor,
+        _res: &Tensor,
+        _grad_res: &Tensor,
+    ) -> Result<(Option<Tensor>, Option<Tensor>)> {
+        Err(crate::Error::BackwardNotSupported { op: self.name() })
+    }
+}
+
+pub trait CustomOp3 {
+    fn name(&self) -> &'static str;
+
+    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cpu_fwd(
+        &self,
+        s1: &CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+        s3: &CpuStorage,
+        l3: &Layout,
+    ) -> Result<(CpuStorage, Shape)>;
+
+    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cuda_fwd(
+        &self,
+        _: &CudaStorage,
+        _: &Layout,
+        _: &CudaStorage,
+        _: &Layout,
+        _: &CudaStorage,
+        _: &Layout,
+    ) -> Result<(CudaStorage, Shape)> {
+        Err(crate::Error::Cuda(
+            format!("no cuda implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn metal_fwd(
+        &self,
+        _: &MetalStorage,
+        _: &Layout,
+        _: &MetalStorage,
+        _: &Layout,
+        _: &MetalStorage,
+        _: &Layout,
+    ) -> Result<(MetalStorage, Shape)> {
+        Err(crate::Error::Metal(
+            format!("no metal implementation for {}", self.name()).into(),
+        ))
+    }
+
+    fn bwd(
+        &self,
+        _arg1: &Tensor,
+        _arg2: &Tensor,
+        _arg3: &Tensor,
+        _res: &Tensor,
+        _grad_res: &Tensor,
+    ) -> Result<(Option<Tensor>, Option<Tensor>, Option<Tensor>)> {
+        Err(crate::Error::BackwardNotSupported { op: self.name() })
+    }
+}
+
+impl Tensor {
+    /// Applies a unary custom op without backward support
+    pub fn apply_op1_no_bwd<C: CustomOp1>(&self, c: &C) -> Result<Self> {
+        let (storage, shape) = self.storage().apply_op1(self.layout(), c)?;
+        Ok(from_storage(storage, shape, BackpropOp::none(), false))
+    }
+
+    /// Applies a binary custom op without backward support
+    pub fn apply_op2_no_bwd<C: CustomOp2>(&self, rhs: &Self, c: &C) -> Result<Self> {
+        let (storage, shape) =
+            self.storage()
+                .apply_op2(self.layout(), &rhs.storage(), rhs.layout(), c)?;
+        Ok(from_storage(storage, shape, BackpropOp::none(), false))
+    }
+
+    /// Applies a ternary custom op without backward support
+    pub fn apply_op3_no_bwd<C: CustomOp3>(&self, t2: &Self, t3: &Self, c: &C) -> Result<Self> {
+        let (storage, shape) = self.storage().apply_op3(
+            self.layout(),
+            &t2.storage(),
+            t2.layout(),
+            &t3.storage(),
+            t3.layout(),
+            c,
+        )?;
+        Ok(from_storage(storage, shape, BackpropOp::none(), false))
+    }
+
+    /// Applies a unary custom op.
+    pub fn apply_op1_arc(&self, c: Arc<Box<dyn CustomOp1 + Send + Sync>>) -> Result<Self> {
+        let (storage, shape) = self
+            .storage()
+            .apply_op1(self.layout(), c.as_ref().as_ref())?;
+        let op = BackpropOp::new1(self, |s| Op::CustomOp1(s, c.clone()));
+        Ok(from_storage(storage, shape, op, false))
+    }
+
+    pub fn apply_op1<C: 'static + CustomOp1 + Send + Sync>(&self, c: C) -> Result<Self> {
+        self.apply_op1_arc(Arc::new(Box::new(c)))
+    }
+
+    /// Applies a binary custom op.
+    pub fn apply_op2_arc(
+        &self,
+        rhs: &Self,
+        c: Arc<Box<dyn CustomOp2 + Send + Sync>>,
+    ) -> Result<Self> {
+        let (storage, shape) = self.storage().apply_op2(
+            self.layout(),
+            &rhs.storage(),
+            rhs.layout(),
+            c.as_ref().as_ref(),
+        )?;
+        let op = BackpropOp::new2(self, rhs, |t1, t2| Op::CustomOp2(t1, t2, c.clone()));
+        Ok(from_storage(storage, shape, op, false))
+    }
+
+    pub fn apply_op2<C: 'static + CustomOp2 + Send + Sync>(&self, r: &Self, c: C) -> Result<Self> {
+        self.apply_op2_arc(r, Arc::new(Box::new(c)))
+    }
+
+    /// Applies a ternary custom op.
+    pub fn apply_op3_arc(
+        &self,
+        t2: &Self,
+        t3: &Self,
+        c: Arc<Box<dyn CustomOp3 + Send + Sync>>,
+    ) -> Result<Self> {
+        let (storage, shape) = self.storage().apply_op3(
+            self.layout(),
+            &t2.storage(),
+            t2.layout(),
+            &t3.storage(),
+            t3.layout(),
+            c.as_ref().as_ref(),
+        )?;
+        let op = BackpropOp::new3(self, t2, t3, |t1, t2, t3| {
+            Op::CustomOp3(t1, t2, t3, c.clone())
+        });
+        Ok(from_storage(storage, shape, op, false))
+    }
+
+    pub fn apply_op3<C: 'static + CustomOp3 + Send + Sync>(
+        &self,
+        t2: &Self,
+        t3: &Self,
+        c: C,
+    ) -> Result<Self> {
+        self.apply_op3_arc(t2, t3, Arc::new(Box::new(c)))
+    }
+}
+
+// In place ops.
+
+/// Unary ops that can be defined in user-land.
+/// These ops work in place and as such back-prop is unsupported.
+pub trait InplaceOp1 {
+    // Box<dyn> does not support const yet, so use a function to get the name.
+    fn name(&self) -> &'static str;
+
+    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cpu_fwd(&self, storage: &mut CpuStorage, layout: &Layout) -> Result<()>;
+
+    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cuda_fwd(&self, _storage: &mut CudaStorage, _layout: &Layout) -> Result<()> {
+        Err(crate::Error::Cuda(
+            format!("no cuda implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn metal_fwd(&self, _storage: &mut MetalStorage, _layout: &Layout) -> Result<()> {
+        Err(crate::Error::Metal(
+            format!("no metal implementation for {}", self.name()).into(),
+        ))
+    }
+}
+
+pub trait InplaceOp2 {
+    fn name(&self) -> &'static str;
+
+    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cpu_fwd(&self, s1: &mut CpuStorage, l1: &Layout, s2: &CpuStorage, l2: &Layout)
+        -> Result<()>;
+
+    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cuda_fwd(&self, _: &mut CudaStorage, _: &Layout, _: &CudaStorage, _: &Layout) -> Result<()> {
+        Err(crate::Error::Cuda(
+            format!("no cuda implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn metal_fwd(
+        &self,
+        _: &mut MetalStorage,
+        _: &Layout,
+        _: &MetalStorage,
+        _: &Layout,
+    ) -> Result<()> {
+        Err(crate::Error::Metal(
+            format!("no metal implementation for {}", self.name()).into(),
+        ))
+    }
+}
+
+pub trait InplaceOp3 {
+    fn name(&self) -> &'static str;
+
+    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cpu_fwd(
+        &self,
+        s1: &mut CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+        s3: &CpuStorage,
+        l3: &Layout,
+    ) -> Result<()>;
+
+    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn cuda_fwd(
+        &self,
+        _: &mut CudaStorage,
+        _: &Layout,
+        _: &CudaStorage,
+        _: &Layout,
+        _: &CudaStorage,
+        _: &Layout,
+    ) -> Result<()> {
+        Err(crate::Error::Cuda(
+            format!("no cuda implementation for {}", self.name()).into(),
+        ))
+    }
+
+    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
+    /// offsets etc so the associated layout should be used to access it.
+    fn metal_fwd(
+        &self,
+        _: &mut MetalStorage,
+        _: &Layout,
+        _: &MetalStorage,
+        _: &Layout,
+        _: &MetalStorage,
+        _: &Layout,
+    ) -> Result<()> {
+        Err(crate::Error::Metal(
+            format!("no metal implementation for {}", self.name()).into(),
+        ))
+    }
+}
+
+impl Tensor {
+    /// Applies a unary custom op in place.
+    pub fn inplace_op1<C: InplaceOp1>(&self, c: &C) -> Result<()> {
+        self.storage_mut().inplace_op1(self.layout(), c)
+    }
+
+    /// Applies a unary custom op in place (for the first tensor).
+    pub fn inplace_op2<C: InplaceOp2>(&self, rhs: &Self, c: &C) -> Result<()> {
+        self.storage_mut()
+            .inplace_op2(self.layout(), &rhs.storage(), rhs.layout(), c)
+    }
+
+    /// Applies a ternary custom op in place (for the first tensor).
+    pub fn inplace_op3<C: InplaceOp3>(&self, t2: &Self, t3: &Self, c: &C) -> Result<()> {
+        self.storage_mut().inplace_op3(
+            self.layout(),
+            &t2.storage(),
+            t2.layout(),
+            &t3.storage(),
+            t3.layout(),
+            c,
+        )
+    }
+}

candle-core/src/device.rs

@@ -289,17 +289,34 @@ impl Device {
         }
     }
 
+    pub(crate) unsafe fn alloc_uninit(&self, shape: &Shape, dtype: DType) -> Result<Storage> {
+        match self {
+            Device::Cpu => {
+                let storage = CpuDevice.alloc_uninit(shape, dtype)?;
+                Ok(Storage::Cpu(storage))
+            }
+            Device::Cuda(device) => {
+                let storage = device.alloc_uninit(shape, dtype)?;
+                Ok(Storage::Cuda(storage))
+            }
+            Device::Metal(device) => {
+                let storage = device.alloc_uninit(shape, dtype)?;
+                Ok(Storage::Metal(storage))
+            }
+        }
+    }
+
     pub(crate) fn storage<A: NdArray>(&self, array: A) -> Result<Storage> {
         match self {
             Device::Cpu => Ok(Storage::Cpu(array.to_cpu_storage())),
             Device::Cuda(device) => {
                 let storage = array.to_cpu_storage();
-                let storage = device.storage_from_cpu_storage(&storage)?;
+                let storage = device.storage_from_cpu_storage_owned(storage)?;
                 Ok(Storage::Cuda(storage))
             }
             Device::Metal(device) => {
                 let storage = array.to_cpu_storage();
-                let storage = device.storage_from_cpu_storage(&storage)?;
+                let storage = device.storage_from_cpu_storage_owned(storage)?;
                 Ok(Storage::Metal(storage))
             }
         }
@@ -310,14 +327,22 @@ impl Device {
             Device::Cpu => Ok(Storage::Cpu(S::to_cpu_storage_owned(data))),
             Device::Cuda(device) => {
                 let storage = S::to_cpu_storage_owned(data);
-                let storage = device.storage_from_cpu_storage(&storage)?;
+                let storage = device.storage_from_cpu_storage_owned(storage)?;
                 Ok(Storage::Cuda(storage))
             }
             Device::Metal(device) => {
                 let storage = S::to_cpu_storage_owned(data);
-                let storage = device.storage_from_cpu_storage(&storage)?;
+                let storage = device.storage_from_cpu_storage_owned(storage)?;
                 Ok(Storage::Metal(storage))
             }
         }
     }
+
+    pub fn synchronize(&self) -> Result<()> {
+        match self {
+            Self::Cpu => Ok(()),
+            Self::Cuda(d) => d.synchronize(),
+            Self::Metal(d) => d.synchronize(),
+        }
+    }
 }

candle-core/src/dummy_cuda_backend.rs

@@ -210,10 +210,18 @@ impl crate::backend::BackendDevice for CudaDevice {
         Err(Error::NotCompiledWithCudaSupport)
     }
 
+    unsafe fn alloc_uninit(&self, _shape: &Shape, _dtype: DType) -> Result<Self::Storage> {
+        Err(Error::NotCompiledWithCudaSupport)
+    }
+
     fn storage_from_cpu_storage(&self, _: &CpuStorage) -> Result<Self::Storage> {
         Err(Error::NotCompiledWithCudaSupport)
     }
 
+    fn storage_from_cpu_storage_owned(&self, _: CpuStorage) -> Result<Self::Storage> {
+        Err(Error::NotCompiledWithCudaSupport)
+    }
+
     fn rand_uniform(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage> {
         Err(Error::NotCompiledWithCudaSupport)
     }
@@ -221,4 +229,8 @@ impl crate::backend::BackendDevice for CudaDevice {
     fn rand_normal(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage> {
         Err(Error::NotCompiledWithCudaSupport)
     }
+
+    fn synchronize(&self) -> Result<()> {
+        Ok(())
+    }
 }

candle-core/src/dummy_metal_backend.rs

@@ -222,10 +222,18 @@ impl crate::backend::BackendDevice for MetalDevice {
         Err(Error::NotCompiledWithMetalSupport)
     }
 
+    unsafe fn alloc_uninit(&self, _shape: &Shape, _dtype: DType) -> Result<Self::Storage> {
+        Err(Error::NotCompiledWithMetalSupport)
+    }
+
     fn storage_from_cpu_storage(&self, _: &CpuStorage) -> Result<Self::Storage> {
         Err(Error::NotCompiledWithMetalSupport)
     }
 
+    fn storage_from_cpu_storage_owned(&self, _: CpuStorage) -> Result<Self::Storage> {
+        Err(Error::NotCompiledWithMetalSupport)
+    }
+
     fn rand_uniform(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage> {
         Err(Error::NotCompiledWithMetalSupport)
     }
@@ -233,4 +241,8 @@ impl crate::backend::BackendDevice for MetalDevice {
     fn rand_normal(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage> {
         Err(Error::NotCompiledWithMetalSupport)
     }
+
+    fn synchronize(&self) -> Result<()> {
+        Ok(())
+    }
 }

candle-core/src/lib.rs

@@ -14,7 +14,7 @@
 //!
 //! ## Features
 //!
-//! - Simple syntax (looks and like PyTorch)
+//! - Simple syntax (looks and feels like PyTorch)
 //! - CPU and Cuda backends (and M1 support)
 //! - Enable serverless (CPU) small and fast deployments
 //! - Model training
@@ -37,14 +37,13 @@
 mod accelerate;
 pub mod backend;
 pub mod backprop;
-mod conv;
+pub mod conv;
 mod convert;
 pub mod cpu;
 pub mod cpu_backend;
 #[cfg(feature = "cuda")]
 pub mod cuda_backend;
-#[cfg(feature = "cudnn")]
-pub mod cudnn;
+mod custom_op;
 mod device;
 pub mod display;
 mod dtype;
@@ -58,7 +57,7 @@ pub mod metal_backend;
 #[cfg(feature = "mkl")]
 mod mkl;
 pub mod npy;
-mod op;
+pub mod op;
 pub mod pickle;
 pub mod quantized;
 pub mod safetensors;
@@ -72,13 +71,16 @@ pub mod test_utils;
 pub mod utils;
 mod variable;
 
+#[cfg(feature = "cudnn")]
+pub use cuda_backend::cudnn;
+
 pub use cpu_backend::CpuStorage;
+pub use custom_op::{CustomOp1, CustomOp2, CustomOp3, InplaceOp1, InplaceOp2, InplaceOp3};
 pub use device::{Device, DeviceLocation, NdArray};
-pub use dtype::{DType, FloatDType, IntDType, WithDType};
+pub use dtype::{DType, DTypeParseError, FloatDType, IntDType, WithDType};
 pub use error::{Error, Result};
 pub use indexer::IndexOp;
 pub use layout::Layout;
-pub use op::{CustomOp1, CustomOp2, CustomOp3};
 pub use shape::{Shape, D};
 pub use storage::Storage;
 pub use strided_index::{StridedBlocks, StridedIndex};

candle-core/src/metal_backend/device.rs

@@ -0,0 +1,287 @@
+use crate::{DType, Result};
+use candle_metal_kernels::Kernels;
+use metal::{Buffer, CommandBuffer, CommandQueue, MTLResourceOptions, NSUInteger};
+use std::collections::HashMap;
+use std::ffi::c_void;
+use std::path::Path;
+use std::sync::{Arc, Mutex, RwLock, RwLockWriteGuard};
+
+use super::MetalError;
+
+/// Unique identifier for cuda devices.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub struct DeviceId(usize);
+
+impl DeviceId {
+    pub(crate) fn new() -> Self {
+        // https://users.rust-lang.org/t/idiomatic-rust-way-to-generate-unique-id/33805
+        use std::sync::atomic;
+        static COUNTER: atomic::AtomicUsize = atomic::AtomicUsize::new(1);
+        Self(COUNTER.fetch_add(1, atomic::Ordering::Relaxed))
+    }
+}
+
+type BufferMap = HashMap<(NSUInteger, MTLResourceOptions), Vec<Arc<Buffer>>>;
+type AllocatedBuffers = Arc<RwLock<BufferMap>>;
+
+#[derive(Clone)]
+pub struct MetalDevice {
+    /// Unique identifier, the registryID is not sufficient as it identifies the GPU rather than
+    /// the device itself.
+    pub(crate) id: DeviceId,
+
+    /// Raw metal device: <https://developer.apple.com/documentation/metal/mtldevice?language=objc>
+    pub(crate) device: metal::Device,
+
+    /// Single command queue for the entire device.
+    pub(crate) command_queue: CommandQueue,
+    /// One command buffer at a time.
+    /// The scheduler works by allowing multiple
+    /// [ComputeCommandEncoder](https://developer.apple.com/documentation/metal/mtlcomputecommandencoder?language=objc)
+    /// on a single command buffer. Using a single command buffer would be fastest on the GPU but
+    /// prevents overlapping of CPU and GPU commands (because command buffer needs to be committed
+    /// to start to work).
+    /// Despite what the documentation says, command buffers are NOT ordered. They are ordered
+    /// for their START time, but there's no guarantee that command buffer1 will finish before
+    /// command buffer2 starts (or there are metal bugs there)
+    pub(crate) command_buffer: Arc<RwLock<CommandBuffer>>,
+    /// Keeps track of the current amount of compute command encoders on the current
+    /// command buffer
+    /// Arc, RwLock because of the interior mutability.
+    pub(crate) command_buffer_index: Arc<RwLock<usize>>,
+    /// The maximum amount of [compute command encoder](https://developer.apple.com/documentation/metal/mtlcomputecommandencoder?language=objc) per [command buffer](https://developer.apple.com/documentation/metal/mtlcommandbuffer?language=objc)
+    pub(crate) compute_per_buffer: usize,
+    /// Simple keeper struct to keep track of the already compiled kernels so we can reuse them.
+    /// Heavily used by [`candle_metal_kernels`]
+    pub(crate) kernels: Arc<Kernels>,
+    /// Simple allocator struct.
+    /// The buffers are stored in size buckets since ML tends to use similar shapes over and over.
+    /// We store the buffers in [`Arc`] because it's much faster than Obj-c internal ref counting
+    /// (could be linked to FFI communication overhead).
+    ///
+    /// Whenever a buffer has a strong_count==1, we can reuse it, it means it was dropped in the
+    /// graph calculation, and only we the allocator kept a reference to it, therefore it's free
+    /// to be reused. However, in order for this to work, we need to guarantee the order of
+    /// operation, so that this buffer is not being used by another kernel at the same time.
+    /// Arc is the CPU reference count, it doesn't mean anything on the GPU side of things.
+    ///
+    /// Whenever we actually allocate a new buffer, we make a full sweep to clean up unused buffers
+    /// (strong_count = 1).
+    pub(crate) buffers: AllocatedBuffers,
+    /// Seed for random number generation.
+    pub(crate) seed: Arc<Mutex<Buffer>>,
+}
+
+impl std::fmt::Debug for MetalDevice {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "MetalDevice({:?})", self.id)
+    }
+}
+
+impl std::ops::Deref for MetalDevice {
+    type Target = metal::DeviceRef;
+
+    fn deref(&self) -> &Self::Target {
+        &self.device
+    }
+}
+
+impl MetalDevice {
+    pub fn id(&self) -> DeviceId {
+        self.id
+    }
+
+    pub fn metal_device(&self) -> &metal::Device {
+        &self.device
+    }
+
+    pub fn command_queue(&self) -> &CommandQueue {
+        &self.command_queue
+    }
+
+    pub fn command_buffer(&self) -> Result<CommandBuffer> {
+        let mut command_buffer_lock = self.command_buffer.try_write().map_err(MetalError::from)?;
+        let mut command_buffer = command_buffer_lock.to_owned();
+        let mut index = self
+            .command_buffer_index
+            .try_write()
+            .map_err(MetalError::from)?;
+        if *index > self.compute_per_buffer {
+            command_buffer.commit();
+            command_buffer = self.command_queue.new_command_buffer().to_owned();
+            *command_buffer_lock = command_buffer.clone();
+            *index = 0;
+
+            self.drop_unused_buffers()?;
+        }
+        *index += 1;
+        Ok(command_buffer)
+    }
+
+    pub fn wait_until_completed(&self) -> Result<()> {
+        let mut command_buffer = self.command_buffer.try_write().map_err(MetalError::from)?;
+        match command_buffer.status() {
+            metal::MTLCommandBufferStatus::Committed
+            | metal::MTLCommandBufferStatus::Scheduled
+            | metal::MTLCommandBufferStatus::Completed => {
+                panic!("Already committed");
+            }
+            _ => {}
+        }
+        command_buffer.commit();
+        command_buffer.wait_until_completed();
+        *command_buffer = self.command_queue.new_command_buffer().to_owned();
+
+        Ok(())
+    }
+
+    pub fn kernels(&self) -> &Kernels {
+        &self.kernels
+    }
+
+    pub fn device(&self) -> &metal::Device {
+        &self.device
+    }
+
+    /// Creates a new buffer (not necessarily zeroed).
+    /// The buffer is [MTLPrivate](https://developer.apple.com/documentation/metal/mtlstoragemode)
+    /// This means the buffer data cannot be read on the CPU directly.
+    ///
+    /// [`name`] is only used to keep track of the resource origin in case of bugs
+    pub fn new_buffer(
+        &self,
+        element_count: usize,
+        dtype: DType,
+        name: &str,
+    ) -> Result<Arc<Buffer>> {
+        let size = (element_count * dtype.size_in_bytes()) as NSUInteger;
+        self.allocate_buffer(size, MTLResourceOptions::StorageModePrivate, name)
+    }
+
+    /// Creates a new buffer (not necessarily zeroed).
+    /// The buffer is [MTLManaged](https://developer.apple.com/documentation/metal/mtlstoragemode)
+    /// This means the buffer can be read on the CPU but will require manual
+    /// synchronization when the CPU memory is modified
+    /// Used as a bridge to gather data back from the GPU
+    pub fn new_buffer_managed(&self, size: NSUInteger) -> Result<Arc<Buffer>> {
+        self.allocate_buffer(size, MTLResourceOptions::StorageModeManaged, "managed")
+    }
+
+    /// Creates a new buffer from data.
+    /// The buffer is [MTLManaged](https://developer.apple.com/documentation/metal/mtlstoragemode)
+    ///
+    /// Does not require synchronization, as [newBufferWithBytes](https://developer.apple.com/documentation/metal/mtldevice/1433429-newbufferwithbytes)
+    /// allocates the buffer and copies over the existing data before returning the MTLBuffer.
+    pub fn new_buffer_with_data<T>(&self, data: &[T]) -> Result<Arc<Buffer>> {
+        let size = core::mem::size_of_val(data) as NSUInteger;
+        let new_buffer = self.device.new_buffer_with_data(
+            data.as_ptr() as *const c_void,
+            size,
+            MTLResourceOptions::StorageModeManaged,
+        );
+        let mut buffers = self.buffers.try_write().map_err(MetalError::from)?;
+        let subbuffers = buffers
+            .entry((size, MTLResourceOptions::StorageModeManaged))
+            .or_insert(vec![]);
+
+        let new_buffer = Arc::new(new_buffer);
+        subbuffers.push(new_buffer.clone());
+        Ok(new_buffer)
+    }
+
+    pub fn allocate_zeros(&self, size_in_bytes: usize) -> Result<Arc<Buffer>> {
+        let buffer = self.allocate_buffer(
+            size_in_bytes as NSUInteger,
+            MTLResourceOptions::StorageModePrivate,
+            "allocate_zeros",
+        )?;
+        let command_buffer = self.command_buffer()?;
+        command_buffer.set_label("zeros");
+        let blit = command_buffer.new_blit_command_encoder();
+        blit.fill_buffer(
+            &buffer,
+            metal::NSRange {
+                location: 0,
+                length: buffer.length(),
+            },
+            0,
+        );
+        blit.end_encoding();
+        Ok(buffer)
+    }
+
+    fn find_available_buffer(
+        &self,
+        size: NSUInteger,
+        option: MTLResourceOptions,
+        buffers: &RwLockWriteGuard<BufferMap>,
+    ) -> Option<Arc<Buffer>> {
+        let mut best_buffer: Option<&Arc<Buffer>> = None;
+        let mut best_buffer_size: NSUInteger = NSUInteger::MAX;
+        for ((buffer_size, buffer_option), subbuffers) in buffers.iter() {
+            if buffer_size >= &size && buffer_size < &best_buffer_size && buffer_option == &option {
+                for sub in subbuffers {
+                    if Arc::strong_count(sub) == 1 {
+                        best_buffer = Some(sub);
+                        best_buffer_size = *buffer_size;
+                    }
+                }
+            }
+        }
+        best_buffer.cloned()
+    }
+
+    fn drop_unused_buffers(&self) -> Result<()> {
+        let mut buffers = self.buffers.try_write().map_err(MetalError::from)?;
+        for subbuffers in buffers.values_mut() {
+            let newbuffers = subbuffers
+                .iter()
+                .filter(|s| Arc::strong_count(*s) > 1)
+                .map(Arc::clone)
+                .collect();
+            *subbuffers = newbuffers;
+        }
+        Ok(())
+    }
+
+    /// The critical allocator algorithm
+    fn allocate_buffer(
+        &self,
+        size: NSUInteger,
+        option: MTLResourceOptions,
+        _name: &str,
+    ) -> Result<Arc<Buffer>> {
+        let mut buffers = self.buffers.try_write().map_err(MetalError::from)?;
+        if let Some(b) = self.find_available_buffer(size, option, &buffers) {
+            // Cloning also ensures we increment the strong count
+            return Ok(b.clone());
+        }
+
+        let size = buf_size(size);
+        let subbuffers = buffers.entry((size, option)).or_insert(vec![]);
+
+        let new_buffer = self.device.new_buffer(size as NSUInteger, option);
+        let new_buffer = Arc::new(new_buffer);
+        subbuffers.push(new_buffer.clone());
+
+        Ok(new_buffer)
+    }
+
+    /// Create a metal GPU capture trace on [`path`].
+    pub fn capture<P: AsRef<Path>>(&self, path: P) -> Result<()> {
+        let capture = metal::CaptureManager::shared();
+        let descriptor = metal::CaptureDescriptor::new();
+        descriptor.set_destination(metal::MTLCaptureDestination::GpuTraceDocument);
+        descriptor.set_capture_device(self);
+        descriptor.set_output_url(path);
+
+        capture
+            .start_capture(&descriptor)
+            .map_err(MetalError::from)?;
+        Ok(())
+    }
+}
+
+fn buf_size(size: NSUInteger) -> NSUInteger {
+    size.saturating_sub(1).next_power_of_two() as NSUInteger
+}

candle-core/src/op.rs

@@ -1,5 +1,5 @@
 #![allow(clippy::redundant_closure_call)]
-use crate::{CpuStorage, CudaStorage, Layout, MetalStorage, Result, Shape, Tensor};
+use crate::Tensor;
 use half::{bf16, f16};
 use num_traits::float::Float;
 
@@ -66,6 +66,7 @@ pub enum UnaryOp {
     Floor,
     Ceil,
     Round,
+    Sign,
 }
 
 #[derive(Clone)]
@@ -161,168 +162,23 @@ pub enum Op {
     Permute(Tensor, Vec<usize>),
     Elu(Tensor, f64),
     Powf(Tensor, f64),
-    CustomOp1(Tensor, std::sync::Arc<Box<dyn CustomOp1 + Send + Sync>>),
+    CustomOp1(
+        Tensor,
+        std::sync::Arc<Box<dyn crate::CustomOp1 + Send + Sync>>,
+    ),
     CustomOp2(
         Tensor,
         Tensor,
-        std::sync::Arc<Box<dyn CustomOp2 + Send + Sync>>,
+        std::sync::Arc<Box<dyn crate::CustomOp2 + Send + Sync>>,
     ),
     CustomOp3(
         Tensor,
         Tensor,
         Tensor,
-        std::sync::Arc<Box<dyn CustomOp3 + Send + Sync>>,
+        std::sync::Arc<Box<dyn crate::CustomOp3 + Send + Sync>>,
     ),
 }
 
-/// Unary ops that can be defined in user-land.
-pub trait CustomOp1 {
-    // Box<dyn> does not support const yet, so use a function to get the name.
-    fn name(&self) -> &'static str;
-
-    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn cpu_fwd(&self, storage: &CpuStorage, layout: &Layout) -> Result<(CpuStorage, Shape)>;
-
-    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn cuda_fwd(&self, _storage: &CudaStorage, _layout: &Layout) -> Result<(CudaStorage, Shape)> {
-        Err(crate::Error::Cuda(
-            format!("no cuda implementation for {}", self.name()).into(),
-        ))
-    }
-
-    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn metal_fwd(
-        &self,
-        _storage: &MetalStorage,
-        _layout: &Layout,
-    ) -> Result<(MetalStorage, Shape)> {
-        Err(crate::Error::Metal(
-            format!("no metal implementation for {}", self.name()).into(),
-        ))
-    }
-
-    /// This function takes as argument the argument `arg` used in the forward pass, the result
-    /// produced by the forward operation `res` and the gradient of the result `grad_res`.
-    /// The function should return the gradient of the argument.
-    fn bwd(&self, _arg: &Tensor, _res: &Tensor, _grad_res: &Tensor) -> Result<Option<Tensor>> {
-        Err(crate::Error::BackwardNotSupported { op: self.name() })
-    }
-}
-
-pub trait CustomOp2 {
-    fn name(&self) -> &'static str;
-
-    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn cpu_fwd(
-        &self,
-        s1: &CpuStorage,
-        l1: &Layout,
-        s2: &CpuStorage,
-        l2: &Layout,
-    ) -> Result<(CpuStorage, Shape)>;
-
-    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn cuda_fwd(
-        &self,
-        _: &CudaStorage,
-        _: &Layout,
-        _: &CudaStorage,
-        _: &Layout,
-    ) -> Result<(CudaStorage, Shape)> {
-        Err(crate::Error::Cuda(
-            format!("no cuda implementation for {}", self.name()).into(),
-        ))
-    }
-
-    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn metal_fwd(
-        &self,
-        _: &MetalStorage,
-        _: &Layout,
-        _: &MetalStorage,
-        _: &Layout,
-    ) -> Result<(MetalStorage, Shape)> {
-        Err(crate::Error::Metal(
-            format!("no metal implementation for {}", self.name()).into(),
-        ))
-    }
-
-    fn bwd(
-        &self,
-        _arg1: &Tensor,
-        _arg2: &Tensor,
-        _res: &Tensor,
-        _grad_res: &Tensor,
-    ) -> Result<(Option<Tensor>, Option<Tensor>)> {
-        Err(crate::Error::BackwardNotSupported { op: self.name() })
-    }
-}
-
-pub trait CustomOp3 {
-    fn name(&self) -> &'static str;
-
-    /// The forward pass, as run on a cpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn cpu_fwd(
-        &self,
-        s1: &CpuStorage,
-        l1: &Layout,
-        s2: &CpuStorage,
-        l2: &Layout,
-        s3: &CpuStorage,
-        l3: &Layout,
-    ) -> Result<(CpuStorage, Shape)>;
-
-    /// The forward pass, as run on a gpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn cuda_fwd(
-        &self,
-        _: &CudaStorage,
-        _: &Layout,
-        _: &CudaStorage,
-        _: &Layout,
-        _: &CudaStorage,
-        _: &Layout,
-    ) -> Result<(CudaStorage, Shape)> {
-        Err(crate::Error::Cuda(
-            format!("no cuda implementation for {}", self.name()).into(),
-        ))
-    }
-
-    /// The forward pass, as run on a metal gpu device. Note that the storage can use arbitrary strides,
-    /// offsets etc so the associated layout should be used to access it.
-    fn metal_fwd(
-        &self,
-        _: &MetalStorage,
-        _: &Layout,
-        _: &MetalStorage,
-        _: &Layout,
-        _: &MetalStorage,
-        _: &Layout,
-    ) -> Result<(MetalStorage, Shape)> {
-        Err(crate::Error::Metal(
-            format!("no metal implementation for {}", self.name()).into(),
-        ))
-    }
-
-    fn bwd(
-        &self,
-        _arg1: &Tensor,
-        _arg2: &Tensor,
-        _arg3: &Tensor,
-        _res: &Tensor,
-        _grad_res: &Tensor,
-    ) -> Result<(Option<Tensor>, Option<Tensor>, Option<Tensor>)> {
-        Err(crate::Error::BackwardNotSupported { op: self.name() })
-    }
-}
-
 pub trait UnaryOpT {
     const NAME: &'static str;
     const KERNEL: &'static str;
@@ -399,6 +255,7 @@ pub(crate) struct Tanh;
 pub(crate) struct Floor;
 pub(crate) struct Ceil;
 pub(crate) struct Round;
+pub(crate) struct Sign;
 
 macro_rules! bin_op {
     ($op:ident, $name: literal, $e: expr, $f32_vec: ident, $f64_vec: ident) => {
@@ -602,6 +459,13 @@ unary_op!(Recip, "recip", v, v.recip());
 unary_op!(Sqr, "sqr", v, v * v, vs_sqr, vd_sqr);
 unary_op!(Sqrt, "sqrt", v, v.sqrt(), vs_sqrt, vd_sqrt);
 
+// Hardcode the value for sqrt(2/pi)
+// https://github.com/huggingface/candle/issues/1982
+#[allow(clippy::excessive_precision)]
+const SQRT_TWO_OVER_PI_F32: f32 = 0.79788456080286535587989211986876373;
+#[allow(clippy::excessive_precision)]
+const SQRT_TWO_OVER_PI_F64: f64 = 0.79788456080286535587989211986876373;
+
 /// Tanh based approximation of the `gelu` operation
 /// GeluErf is the more precise one.
 /// <https://en.wikipedia.org/wiki/Activation_function#Comparison_of_activation_functions>
@@ -614,7 +478,7 @@ impl UnaryOpT for Gelu {
             * v
             * (bf16::ONE
                 + bf16::tanh(
-                    (bf16::from_f32_const(2.0) / bf16::PI).sqrt()
+                    bf16::from_f32_const(SQRT_TWO_OVER_PI_F32)
                         * v
                         * (bf16::ONE + bf16::from_f32_const(0.044715) * v * v),
                 ))
@@ -625,22 +489,18 @@ impl UnaryOpT for Gelu {
             * v
             * (f16::ONE
                 + f16::tanh(
-                    (f16::from_f32_const(2.0) / f16::PI).sqrt()
+                    f16::from_f32_const(SQRT_TWO_OVER_PI_F32)
                         * v
                         * (f16::ONE + f16::from_f32_const(0.044715) * v * v),
                 ))
     }
     #[inline(always)]
     fn f32(v: f32) -> f32 {
-        0.5 * v
-            * (1.0
-                + f32::tanh((2.0f32 / std::f32::consts::PI).sqrt() * v * (1.0 + 0.044715 * v * v)))
+        0.5 * v * (1.0 + f32::tanh(SQRT_TWO_OVER_PI_F32 * v * (1.0 + 0.044715 * v * v)))
     }
     #[inline(always)]
     fn f64(v: f64) -> f64 {
-        0.5 * v
-            * (1.0
-                + f64::tanh((2.0f64 / std::f64::consts::PI).sqrt() * v * (1.0 + 0.044715 * v * v)))
+        0.5 * v * (1.0 + f64::tanh(SQRT_TWO_OVER_PI_F64 * v * (1.0 + 0.044715 * v * v)))
     }
     #[inline(always)]
     fn u8(_: u8) -> u8 {
@@ -1067,3 +927,37 @@ impl std::ops::Deref for BackpropOp {
         &self.0
     }
 }
+
+impl UnaryOpT for Sign {
+    const NAME: &'static str = "sign";
+    const KERNEL: &'static str = "usign";
+    const V: Self = Sign;
+    #[inline(always)]
+    fn bf16(v: bf16) -> bf16 {
+        bf16::from((v > bf16::ZERO) as i8) - bf16::from((v < bf16::ZERO) as i8)
+    }
+    #[inline(always)]
+    fn f16(v: f16) -> f16 {
+        f16::from((v > f16::ZERO) as i8) - f16::from((v < f16::ZERO) as i8)
+    }
+    #[inline(always)]
+    fn f32(v: f32) -> f32 {
+        f32::from(v > 0.) - f32::from(v < 0.)
+    }
+    #[inline(always)]
+    fn f64(v: f64) -> f64 {
+        f64::from(v > 0.) - f64::from(v < 0.)
+    }
+    #[inline(always)]
+    fn u8(v: u8) -> u8 {
+        u8::min(1, v)
+    }
+    #[inline(always)]
+    fn u32(v: u32) -> u32 {
+        u32::min(1, v)
+    }
+    #[inline(always)]
+    fn i64(v: i64) -> i64 {
+        (v > 0) as i64 - (v < 0) as i64
+    }
+}

candle-core/src/quantized/cuda.rs

@@ -1,22 +1,63 @@
 use super::{GgmlDType, QStorage};
+use crate::quantized::k_quants::GgmlType;
 use crate::{backend::BackendDevice, cuda_backend::WrapErr};
 use crate::{CudaDevice, CudaStorage, Result};
 
-use cudarc::driver::{CudaSlice, DeviceSlice};
+use cudarc::driver::{CudaSlice, CudaView, DeviceSlice};
 
+#[derive(Clone, Debug)]
 pub struct QCudaStorage {
     data: CudaSlice<u8>,
     dtype: GgmlDType,
     device: CudaDevice,
 }
 
+static FORCE_DMMV: std::sync::atomic::AtomicBool = std::sync::atomic::AtomicBool::new(false);
+
+pub fn set_force_dmmv(f: bool) {
+    FORCE_DMMV.store(f, std::sync::atomic::Ordering::Relaxed)
+}
+
 pub const WARP_SIZE: usize = 32;
 pub const MMQ_X_Q4_0_AMPERE: usize = 4;
 pub const MMQ_Y_Q4_0_AMPERE: usize = 32;
 pub const NWARPS_Q4_0_AMPERE: usize = 4;
 pub const GGML_CUDA_MMV_X: usize = 32;
 pub const GGML_CUDA_MMV_Y: usize = 1;
+pub const CUDA_QUANTIZE_BLOCK_SIZE: usize = 256;
 pub const CUDA_DEQUANTIZE_BLOCK_SIZE: usize = 256;
+pub const MATRIX_ROW_PADDING: usize = 512;
+
+fn ceil_div(p: usize, q: usize) -> usize {
+    (p + q - 1) / q
+}
+
+fn pad(p: usize, q: usize) -> usize {
+    ceil_div(p, q) * q
+}
+
+fn quantize_q8_1(
+    src: &CudaView<f32>,
+    dst: &mut CudaSlice<u8>,
+    elem_count: usize,
+    ky: usize,
+    dev: &CudaDevice,
+) -> Result<()> {
+    use cudarc::driver::LaunchAsync;
+
+    let kx = elem_count;
+    let kx_padded = pad(kx, MATRIX_ROW_PADDING);
+    let num_blocks = ceil_div(kx_padded, CUDA_QUANTIZE_BLOCK_SIZE);
+    let func = dev.get_or_load_func("quantize_q8_1", candle_kernels::QUANTIZED)?;
+    let cfg = cudarc::driver::LaunchConfig {
+        grid_dim: (num_blocks as u32, ky as u32, 1),
+        block_dim: (CUDA_QUANTIZE_BLOCK_SIZE as u32, 1, 1),
+        shared_mem_bytes: 0,
+    };
+    let params = (src, dst, kx as i32, kx_padded as i32);
+    unsafe { func.launch(cfg, params) }.w()?;
+    Ok(())
+}
 
 fn dequantize(
     data: &CudaSlice<u8>,
@@ -30,26 +71,18 @@ fn dequantize(
     let (kernel_name, is_k, block_dim, num_blocks) = match dtype {
         GgmlDType::Q4_0 => ("dequantize_block_q4_0", false, 32, nb),
         GgmlDType::Q4_1 => ("dequantize_block_q4_1", false, 32, nb),
-        GgmlDType::Q5_0 => {
-            let nb = (elem_count + 2 * CUDA_DEQUANTIZE_BLOCK_SIZE - 1)
-                / (2 * CUDA_DEQUANTIZE_BLOCK_SIZE);
-            (
-                "dequantize_block_q5_0",
-                false,
-                CUDA_DEQUANTIZE_BLOCK_SIZE,
-                nb,
-            )
-        }
-        GgmlDType::Q5_1 => {
-            let nb = (elem_count + 2 * CUDA_DEQUANTIZE_BLOCK_SIZE - 1)
-                / (2 * CUDA_DEQUANTIZE_BLOCK_SIZE);
-            (
-                "dequantize_block_q5_1",
-                false,
-                CUDA_DEQUANTIZE_BLOCK_SIZE,
-                nb,
-            )
-        }
+        GgmlDType::Q5_0 => (
+            "dequantize_block_q5_0",
+            false,
+            CUDA_DEQUANTIZE_BLOCK_SIZE,
+            ceil_div(elem_count, 2 * CUDA_DEQUANTIZE_BLOCK_SIZE),
+        ),
+        GgmlDType::Q5_1 => (
+            "dequantize_block_q5_1",
+            false,
+            CUDA_DEQUANTIZE_BLOCK_SIZE,
+            ceil_div(elem_count, 2 * CUDA_DEQUANTIZE_BLOCK_SIZE),
+        ),
         GgmlDType::Q8_0 => ("dequantize_block_q8_0", false, 32, nb),
         GgmlDType::Q2K => ("dequantize_block_q2_K", true, 64, nb),
         GgmlDType::Q3K => ("dequantize_block_q3_K", true, 64, nb),
@@ -60,7 +93,7 @@ fn dequantize(
         _ => crate::bail!("unsupported dtype for dequantize {dtype:?}"),
     };
     let func = dev.get_or_load_func(kernel_name, candle_kernels::QUANTIZED)?;
-    let dst = dev.alloc_zeros::<f32>(elem_count).w()?;
+    let dst = unsafe { dev.alloc::<f32>(elem_count).w()? };
     // See e.g.
     // https://github.com/ggerganov/llama.cpp/blob/cbbd1efa06f8c09f9dff58ff9d9af509cc4c152b/ggml-cuda.cu#L7270
     let cfg = cudarc::driver::LaunchConfig {
@@ -83,9 +116,9 @@ fn dequantize(
     Ok(CudaStorage::wrap_cuda_slice(dst, dev.clone()))
 }
 
-fn dequantize_mut_mal_vec(
+fn dequantize_mul_mat_vec(
     data: &CudaSlice<u8>,
-    y: &cudarc::driver::CudaView<f32>,
+    y: &CudaView<f32>,
     dtype: GgmlDType,
     ncols: usize,
     nrows: usize,
@@ -93,6 +126,13 @@ fn dequantize_mut_mal_vec(
 ) -> Result<CudaStorage> {
     use cudarc::driver::LaunchAsync;
 
+    let data_elems = data.len() / dtype.type_size() * dtype.block_size();
+    if data_elems < ncols * nrows {
+        crate::bail!("unexpected data size {}, ncols {ncols} {nrows}", data_elems)
+    }
+    if y.len() != ncols {
+        crate::bail!("unexpected y size {}, ncols {ncols} {nrows}", y.len())
+    }
     let kernel_name = match dtype {
         GgmlDType::Q4_0 => "dequantize_mul_mat_vec_q4_0_cuda",
         GgmlDType::Q4_1 => "dequantize_mul_mat_vec_q4_1_cuda",
@@ -107,8 +147,8 @@ fn dequantize_mut_mal_vec(
         _ => crate::bail!("unsupported dtype for quantized matmul {dtype:?}"),
     };
     let func = dev.get_or_load_func(kernel_name, candle_kernels::QUANTIZED)?;
-    let dst = dev.alloc_zeros::<f32>(nrows).w()?;
-    let block_num_y = (nrows + GGML_CUDA_MMV_Y - 1) / GGML_CUDA_MMV_Y;
+    let dst = unsafe { dev.alloc::<f32>(nrows).w()? };
+    let block_num_y = ceil_div(nrows, GGML_CUDA_MMV_Y);
     let cfg = cudarc::driver::LaunchConfig {
         grid_dim: (block_num_y as u32, 1, 1),
         block_dim: (WARP_SIZE as u32, GGML_CUDA_MMV_Y as u32, 1),
@@ -120,9 +160,149 @@ fn dequantize_mut_mal_vec(
     Ok(CudaStorage::wrap_cuda_slice(dst, dev.clone()))
 }
 
+fn mul_mat_vec_via_q8_1(
+    data: &CudaSlice<u8>,
+    y: &CudaView<f32>,
+    dtype: GgmlDType,
+    ncols: usize,
+    nrows: usize,
+    b_size: usize,
+    dev: &CudaDevice,
+) -> Result<CudaStorage> {
+    use cudarc::driver::LaunchAsync;
+
+    let data_elems = data.len() / dtype.type_size() * dtype.block_size();
+    if data_elems < ncols * nrows {
+        crate::bail!("unexpected data size {}, ncols {ncols} {nrows}", data_elems)
+    }
+    if y.len() != ncols * b_size {
+        crate::bail!("unexpected y size {}, ncols {ncols} {nrows}", y.len())
+    }
+    if b_size == 0 || b_size > 8 {
+        crate::bail!("only bsize between 1 and 8 are supported, got {b_size}")
+    }
+    // Start by quantizing y
+    let ncols_padded = pad(ncols, MATRIX_ROW_PADDING);
+    let y_size_in_bytes =
+        b_size * ncols_padded * GgmlDType::Q8_1.type_size() / GgmlDType::Q8_1.block_size();
+    let mut y_q8_1 = unsafe { dev.alloc::<u8>(y_size_in_bytes).w()? };
+    quantize_q8_1(y, &mut y_q8_1, ncols, b_size, dev)?;
+
+    let kernel_name = match dtype {
+        GgmlDType::Q4_0 => "mul_mat_vec_q4_0_q8_1_cuda",
+        GgmlDType::Q4_1 => "mul_mat_vec_q4_1_q8_1_cuda",
+        GgmlDType::Q5_0 => "mul_mat_vec_q5_0_q8_1_cuda",
+        GgmlDType::Q5_1 => "mul_mat_vec_q5_1_q8_1_cuda",
+        GgmlDType::Q8_0 => "mul_mat_vec_q8_0_q8_1_cuda",
+        GgmlDType::Q2K => "mul_mat_vec_q2_K_q8_1_cuda",
+        GgmlDType::Q3K => "mul_mat_vec_q3_K_q8_1_cuda",
+        GgmlDType::Q4K => "mul_mat_vec_q4_K_q8_1_cuda",
+        GgmlDType::Q5K => "mul_mat_vec_q5_K_q8_1_cuda",
+        GgmlDType::Q6K => "mul_mat_vec_q6_K_q8_1_cuda",
+        _ => crate::bail!("unsupported dtype for quantized matmul {dtype:?}"),
+    };
+    let kernel_name = format!("{kernel_name}{b_size}");
+    let func = dev.get_or_load_func(&kernel_name, candle_kernels::QUANTIZED)?;
+    let dst = unsafe { dev.alloc::<f32>(nrows * b_size).w()? };
+    // https://github.com/ggerganov/llama.cpp/blob/facb8b56f8fd3bb10a693bf0943ae9d69d0828ef/ggml-cuda/mmvq.cu#L98
+    let (nblocks, nwarps) = match b_size {
+        1 => (nrows as u32, 4),
+        2..=4 => ((nrows as u32 + 1) / 2, 4),
+        5..=8 => ((nrows as u32 + 1) / 2, 2),
+        _ => crate::bail!("unexpected bsize {b_size}"),
+    };
+    let cfg = cudarc::driver::LaunchConfig {
+        grid_dim: (nblocks, 1, 1),
+        block_dim: (WARP_SIZE as u32, nwarps, 1),
+        shared_mem_bytes: 0,
+    };
+
+    let params = (
+        data,
+        &y_q8_1,
+        &dst,
+        /* ncols_x */ ncols as i32,
+        /* nrows_x */ nrows as i32,
+        /* nrows_y */ ncols_padded as i32,
+        /* nrows_dst */ nrows as i32,
+    );
+    unsafe { func.launch(cfg, params) }.w()?;
+    Ok(CudaStorage::wrap_cuda_slice(dst, dev.clone()))
+}
+
+#[allow(clippy::too_many_arguments)]
+fn mul_mat_via_q8_1(
+    data: &CudaSlice<u8>,
+    y: &CudaView<f32>,
+    dtype: GgmlDType,
+    x_rows: usize,
+    x_cols: usize,
+    y_rows: usize,
+    y_cols: usize,
+    dev: &CudaDevice,
+) -> Result<CudaStorage> {
+    use cudarc::driver::LaunchAsync;
+
+    let data_elems = data.len() / dtype.type_size() * dtype.block_size();
+    if data_elems < x_rows * x_cols {
+        crate::bail!("unexpected lhs size {}, {x_rows} {x_cols}", data_elems)
+    }
+    if y.len() != y_rows * y_cols {
+        crate::bail!("unexpected y size {}, {y_rows} {y_cols}", y.len())
+    }
+    if x_cols != y_rows {
+        crate::bail!("unexpected x/y size {x_rows} {x_cols} {y_rows} {y_cols}")
+    }
+    let k = x_cols;
+    // Start by quantizing y
+    let k_padded = pad(k, MATRIX_ROW_PADDING);
+    let y_size_in_bytes =
+        k_padded * y_rows * GgmlDType::Q8_1.type_size() / GgmlDType::Q8_1.block_size();
+    let mut y_q8_1 = unsafe { dev.alloc::<u8>(y_size_in_bytes).w()? };
+    quantize_q8_1(y, &mut y_q8_1, k, y_cols, dev)?;
+
+    let (kernel_name, mmq_x, mmq_y) = match dtype {
+        GgmlDType::Q4_0 => ("mul_mat_q4_0", 64, 128),
+        GgmlDType::Q4_1 => ("mul_mat_q4_1", 64, 128),
+        GgmlDType::Q5_0 => ("mul_mat_q5_0", 128, 64),
+        GgmlDType::Q5_1 => ("mul_mat_q5_1", 128, 64),
+        GgmlDType::Q8_0 => ("mul_mat_q8_0", 128, 64),
+        GgmlDType::Q2K => ("mul_mat_q2_K", 64, 128),
+        GgmlDType::Q3K => ("mul_mat_q3_K", 128, 128),
+        GgmlDType::Q4K => ("mul_mat_q4_K", 64, 128),
+        GgmlDType::Q5K => ("mul_mat_q5_K", 64, 128),
+        GgmlDType::Q6K => ("mul_mat_q6_K", 64, 64),
+        _ => crate::bail!("unsupported dtype for quantized matmul {dtype:?}"),
+    };
+    let func = dev.get_or_load_func(kernel_name, candle_kernels::QUANTIZED)?;
+    let dst = unsafe { dev.alloc::<f32>(x_rows * y_cols).w()? };
+    let cfg = cudarc::driver::LaunchConfig {
+        grid_dim: (
+            ceil_div(x_rows, mmq_y) as u32,
+            ceil_div(y_cols, mmq_x) as u32,
+            1,
+        ),
+        block_dim: (WARP_SIZE as u32, 4, 1),
+        shared_mem_bytes: 0,
+    };
+
+    let params = (
+        /* vx */ data,
+        /* vy */ &y_q8_1,
+        /* dst */ &dst,
+        /* ncols_x */ x_cols as i32,
+        /* nrows_x */ x_rows as i32,
+        /* ncols_y */ y_cols as i32,
+        /* nrows_y */ k_padded as i32,
+        /* nrows_dst */ x_rows as i32,
+    );
+    unsafe { func.launch(cfg, params) }.w()?;
+    Ok(CudaStorage::wrap_cuda_slice(dst, dev.clone()))
+}
+
 impl QCudaStorage {
     pub fn zeros(device: &CudaDevice, el_count: usize, dtype: GgmlDType) -> Result<Self> {
-        let size_in_bytes = el_count * dtype.type_size() / dtype.block_size();
+        let size_in_bytes = ceil_div(el_count, dtype.block_size()) * dtype.type_size();
         let data = device.alloc_zeros::<u8>(size_in_bytes).w()?;
         Ok(QCudaStorage {
             data,
@@ -140,6 +320,12 @@ impl QCudaStorage {
     }
 
     pub fn dequantize(&self, elem_count: usize) -> Result<CudaStorage> {
+        fn deq<T: GgmlType>(buffer: &[u8], n: usize, dst: &mut [f32]) -> Result<()> {
+            let slice = unsafe { std::slice::from_raw_parts(buffer.as_ptr() as *const T, n) };
+            let vec = slice.to_vec();
+            T::to_float(&vec, dst)
+        }
+
         let fast_kernel = matches!(
             self.dtype,
             GgmlDType::Q4_0
@@ -158,69 +344,25 @@ impl QCudaStorage {
             return dequantize(&self.data, self.dtype, elem_count, self.device());
         }
         // Run the dequantization on cpu.
-        use crate::quantized::k_quants::GgmlType;
 
         let buffer = self.device.dtoh_sync_copy(&self.data).w()?;
         let mut out = vec![0.0; elem_count];
         let block_len = elem_count / self.dtype.block_size();
         match self.dtype {
-            GgmlDType::F32 => {
-                let slice =
-                    unsafe { std::slice::from_raw_parts(buffer.as_ptr() as *const f32, block_len) };
-                out.copy_from_slice(slice)
-            }
-            GgmlDType::F16 => {
-                let vec: Vec<half::f16> = read_to_vec(&buffer, block_len);
-                half::f16::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q4_0 => {
-                let vec: Vec<crate::quantized::BlockQ4_0> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ4_0::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q4_1 => {
-                let vec: Vec<crate::quantized::BlockQ4_1> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ4_1::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q5_0 => {
-                let vec: Vec<crate::quantized::BlockQ5_0> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ5_0::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q5_1 => {
-                let vec: Vec<crate::quantized::BlockQ5_1> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ5_1::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q8_0 => {
-                let vec: Vec<crate::quantized::BlockQ8_0> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ8_0::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q8_1 => {
-                let vec: Vec<crate::quantized::BlockQ8_1> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ8_1::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q2K => {
-                let vec: Vec<crate::quantized::BlockQ2K> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ2K::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q3K => {
-                let vec: Vec<crate::quantized::BlockQ3K> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ3K::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q4K => {
-                let vec: Vec<crate::quantized::BlockQ4K> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ4K::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q5K => {
-                let vec: Vec<crate::quantized::BlockQ5K> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ5K::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q6K => {
-                let vec: Vec<crate::quantized::BlockQ6K> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ6K::to_float(&vec, &mut out)?;
-            }
-            GgmlDType::Q8K => {
-                let vec: Vec<crate::quantized::BlockQ8K> = read_to_vec(&buffer, block_len);
-                crate::quantized::BlockQ8K::to_float(&vec, &mut out)?;
-            }
+            GgmlDType::F32 => deq::<f32>(&buffer, block_len, &mut out)?,
+            GgmlDType::F16 => deq::<half::f16>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q4_0 => deq::<crate::quantized::BlockQ4_0>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q4_1 => deq::<crate::quantized::BlockQ4_1>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q5_0 => deq::<crate::quantized::BlockQ5_0>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q5_1 => deq::<crate::quantized::BlockQ5_1>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q8_0 => deq::<crate::quantized::BlockQ8_0>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q8_1 => deq::<crate::quantized::BlockQ8_1>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q2K => deq::<crate::quantized::BlockQ2K>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q3K => deq::<crate::quantized::BlockQ3K>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q4K => deq::<crate::quantized::BlockQ4K>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q5K => deq::<crate::quantized::BlockQ5K>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q6K => deq::<crate::quantized::BlockQ6K>(&buffer, block_len, &mut out)?,
+            GgmlDType::Q8K => deq::<crate::quantized::BlockQ8K>(&buffer, block_len, &mut out)?,
         }
 
         self.device
@@ -255,7 +397,17 @@ impl QCudaStorage {
         storage: &CudaStorage,
         layout: &crate::Layout,
     ) -> Result<(CudaStorage, crate::Shape)> {
-        if matches!(layout.shape().dims(), [1, 1, _] | [1, _]) {
+        let max_bm = if FORCE_DMMV.load(std::sync::atomic::Ordering::Relaxed) {
+            1
+        } else {
+            8
+        };
+        let use_vec_kernel = match layout.shape().dims() {
+            [b, m, _k] => b * m <= max_bm,
+            [b, _k] => *b <= max_bm,
+            _ => false,
+        };
+        if use_vec_kernel {
             self.dequantize_matmul_vec(self_shape, storage, layout)
         } else {
             self.dequantize_matmul(self_shape, storage, layout)
@@ -276,22 +428,31 @@ impl QCudaStorage {
             Some((o1, o2)) => rhs.slice(o1..o2),
             None => Err(crate::Error::RequiresContiguous { op: "dmmv" }.bt())?,
         };
-        let (with_batch, k) = match rhs_l.shape().dims() {
-            [1, 1, k] => (true, k),
-            [1, k] => (false, k),
+        let (b_size, k) = match rhs_l.shape().dims() {
+            [b, m, k] => (b * m, *k),
+            [b, k] => (*b, *k),
             _ => crate::bail!("unexpected rhs shape in dmmv {:?}", rhs_l.shape()),
         };
-        if ncols != *k {
+        if ncols != k {
             crate::bail!("mismatch on matmul dim {self_shape:?} {:?}", rhs_l.shape())
         }
 
-        let out =
-            dequantize_mut_mal_vec(&self.data, &rhs, self.dtype, ncols, nrows, self.device())?;
-        let out_shape = if with_batch {
-            vec![1, 1, nrows]
+        let out = if FORCE_DMMV.load(std::sync::atomic::Ordering::Relaxed) {
+            dequantize_mul_mat_vec(&self.data, &rhs, self.dtype, ncols, nrows, self.device())?
         } else {
-            vec![1, nrows]
+            mul_mat_vec_via_q8_1(
+                &self.data,
+                &rhs,
+                self.dtype,
+                ncols,
+                nrows,
+                b_size,
+                self.device(),
+            )?
         };
+        let mut out_shape = rhs_l.shape().dims().to_vec();
+        out_shape.pop();
+        out_shape.push(nrows);
         Ok((out, out_shape.into()))
     }
 
@@ -312,9 +473,30 @@ impl QCudaStorage {
             crate::bail!("mismatch on matmul dim {self_shape:?} {:?}", layout.shape())
         }
 
-        let data_f32 = self.dequantize(n * k)?;
-        let rhs_l = crate::Layout::new((k, n).into(), vec![1, k], 0).broadcast_as((b, k, n))?;
-        let out = storage.matmul(&data_f32, (b, m, n, k), layout, &rhs_l)?;
+        let out = if FORCE_DMMV.load(std::sync::atomic::Ordering::Relaxed) {
+            let data_f32 = self.dequantize(n * k)?;
+            let rhs_l = crate::Layout::new((k, n).into(), vec![1, k], 0).broadcast_as((b, k, n))?;
+            storage.matmul(&data_f32, (b, m, n, k), layout, &rhs_l)?
+        } else {
+            let storage = storage.as_cuda_slice::<f32>()?;
+            let storage = match layout.contiguous_offsets() {
+                Some((o1, o2)) => storage.slice(o1..o2),
+                None => Err(crate::Error::RequiresContiguous {
+                    op: "quantized-matmul",
+                }
+                .bt())?,
+            };
+            mul_mat_via_q8_1(
+                &self.data,
+                &storage,
+                self.dtype,
+                /* x_rows */ n,
+                /* x_cols */ k,
+                /* y_rows */ k,
+                /* y_cols */ b * m,
+                self.device(),
+            )?
+        };
         let mut out_shape = layout.shape().dims().to_vec();
         out_shape.pop();
         out_shape.push(n);
@@ -322,11 +504,6 @@ impl QCudaStorage {
     }
 }
 
-fn read_to_vec<T: Clone>(buffer: &[u8], n: usize) -> Vec<T> {
-    let slice = unsafe { std::slice::from_raw_parts(buffer.as_ptr() as *const T, n) };
-    slice.to_vec()
-}
-
 pub fn load_quantized<T: super::GgmlType + Send + Sync + 'static>(
     device: &CudaDevice,
     data: &[T],
@@ -341,3 +518,101 @@ pub fn load_quantized<T: super::GgmlType + Send + Sync + 'static>(
         dtype: T::DTYPE,
     }))
 }
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn cuda_quantize_q8_1() -> Result<()> {
+        let dev = CudaDevice::new(0)?;
+        let el = 256;
+        let el_padded = pad(el, MATRIX_ROW_PADDING);
+        let y_size_in_bytes =
+            el_padded * GgmlDType::Q8_1.type_size() / GgmlDType::Q8_1.block_size();
+        let mut y_q8_1 = unsafe { dev.alloc::<u8>(y_size_in_bytes).w()? };
+        let vs: Vec<f32> = (0..el).map(|v| v as f32).collect();
+        let y = dev.htod_sync_copy(&vs).w()?;
+        quantize_q8_1(&y.slice(..), &mut y_q8_1, el, 1, &dev)?;
+        Ok(())
+    }
+
+    #[test]
+    fn cuda_mmv_q8_1() -> Result<()> {
+        let dev = CudaDevice::new(0)?;
+        let ncols = 256;
+        let vs: Vec<f32> = (0..ncols).map(|v| v as f32).collect();
+        let y = dev.htod_sync_copy(&vs).w()?;
+        let mut xs = QCudaStorage::zeros(&dev, ncols, GgmlDType::Q4_0)?;
+        xs.quantize(&CudaStorage::wrap_cuda_slice(y.clone(), dev.clone()))?;
+        let cuda_storage = mul_mat_vec_via_q8_1(
+            &xs.data,
+            &y.slice(..),
+            /* dtype */ GgmlDType::Q4_0,
+            /* ncols */ ncols,
+            /* nrows */ 1,
+            /* b_size */ 1,
+            &dev,
+        )?;
+        let vs = cuda_storage.as_cuda_slice::<f32>()?;
+        let vs = dev.dtoh_sync_copy(&vs.slice(..)).unwrap();
+        assert_eq!(vs.len(), 1);
+        // for n = 255, n.(n+1).(2n+1) / 6 = 5559680
+        // Q8 means 1/256 precision.
+        assert_eq!(vs[0], 5561664.5);
+
+        let cuda_storage = dequantize_mul_mat_vec(
+            &xs.data,
+            &y.slice(..),
+            /* dtype */ GgmlDType::Q4_0,
+            /* ncols */ ncols,
+            /* nrows */ 1,
+            &dev,
+        )?;
+        let vs = cuda_storage.as_cuda_slice::<f32>()?;
+        let vs = dev.dtoh_sync_copy(&vs.slice(..)).unwrap();
+        assert_eq!(vs.len(), 1);
+        assert_eq!(vs[0], 5561851.0);
+        Ok(())
+    }
+
+    #[test]
+    fn cuda_mm_q8_1() -> Result<()> {
+        let dev = CudaDevice::new(0)?;
+        let ncols = 256;
+        let vs: Vec<f32> = (0..ncols * 4).map(|v| v as f32 / 4.).collect();
+        let y = dev.htod_sync_copy(&vs).w()?;
+        let mut xs = QCudaStorage::zeros(&dev, ncols * 4, GgmlDType::Q4_0)?;
+        xs.quantize(&CudaStorage::wrap_cuda_slice(y.clone(), dev.clone()))?;
+        let cuda_storage = mul_mat_via_q8_1(
+            &xs.data,
+            &y.slice(..),
+            /* dtype */ GgmlDType::Q4_0,
+            /* x_rows */ 4,
+            /* x_cols */ ncols,
+            /* y_rows */ ncols,
+            /* y_cols */ 4,
+            &dev,
+        )?;
+        let vs = cuda_storage.as_cuda_slice::<f32>()?;
+        let vs = dev.dtoh_sync_copy(&vs.slice(..)).unwrap();
+
+        /*
+           x = torch.tensor([float(v) for v in range(1024)]).reshape(4, 256)
+           x @ x.t() / 16
+        tensor([[  347480.0000,   869720.0000,  1391960.0000,  1914200.0000],
+                [  869720.0000,  2440536.0000,  4011352.0000,  5582166.5000],
+                [ 1391960.0000,  4011352.0000,  6630742.0000,  9250132.0000],
+                [ 1914200.0000,  5582166.5000,  9250132.0000, 12918099.0000]])
+                */
+        assert_eq!(vs.len(), 16);
+        assert_eq!(vs[0], 347604.0);
+        assert_eq!(vs[1], 888153.06);
+        assert_eq!(vs[4], 869780.7);
+        assert_eq!(vs[5], 2483145.0);
+        assert_eq!(vs[11], 9407368.0);
+        assert_eq!(vs[14], 9470856.0);
+        assert_eq!(vs[15], 13138824.0);
+        Ok(())
+    }
+}

candle-core/src/quantized/metal.rs

@@ -149,8 +149,11 @@ impl QMetalStorage {
         let (n, k) = self_shape.dims2()?;
         let mut dst_shape = src_shape.dims().to_vec();
 
+        // We always use a single batch dimension and stack all the tensors in the batch on the
+        // second dimension as the implementation in candle-metal-kernels doesn't handle batch
+        // properly.
         let (b, m) = match dst_shape.len() {
-            3 => (dst_shape[0], dst_shape[1]),
+            3 => (1, dst_shape[0] * dst_shape[1]),
             2 => (1, dst_shape[0]),
             n => crate::bail!("Invalid rank {n} for quantized matmul metal"),
         };

candle-core/src/shape.rs

@@ -171,7 +171,7 @@ impl Shape {
         }
         let mut acc = 1;
         for (&stride, &dim) in stride.iter().zip(self.0.iter()).rev() {
-            if stride != acc {
+            if dim > 1 && stride != acc {
                 return false;
             }
             acc *= dim;
@@ -186,7 +186,7 @@ impl Shape {
         }
         let mut acc = 1;
         for (&stride, &dim) in stride.iter().zip(self.0.iter()) {
-            if stride != acc {
+            if dim > 1 && stride != acc {
                 return false;
             }
             acc *= dim;

candle-core/src/storage.rs

@@ -1,6 +1,7 @@
 use crate::backend::BackendStorage;
-use crate::op::{self, CmpOp, CustomOp1, CustomOp2, CustomOp3, ReduceOp};
+use crate::op::{self, CmpOp, ReduceOp};
 use crate::{CpuStorage, CudaStorage, DType, Device, Error, Layout, MetalStorage, Result, Shape};
+use crate::{CustomOp1, CustomOp2, CustomOp3, InplaceOp1, InplaceOp2, InplaceOp3};
 
 // We do not want to implement Clone on Storage as cloning may fail because of
 // out of memory. Instead try_clone should be used.
@@ -43,9 +44,19 @@ impl Storage {
     }
 
     pub(crate) fn same_device(&self, rhs: &Self, op: &'static str) -> Result<()> {
-        let lhs = self.device().location();
-        let rhs = rhs.device().location();
-        if lhs != rhs {
+        let lhs_device = self.device();
+        let rhs_device = rhs.device();
+        let lhs = lhs_device.location();
+        let rhs = rhs_device.location();
+        let same_device = if self.device().is_metal() {
+            // On metal, we require the device to be exactly the same rather than
+            // having the same location. In cuda this is not necessary as all CudaDevice on the
+            // same GPU will use the same cuda stream.
+            lhs_device.same_device(&rhs_device)
+        } else {
+            lhs == rhs
+        };
+        if !same_device {
             Err(Error::DeviceMismatchBinaryOp { lhs, rhs, op }.bt())
         } else {
             Ok(())
@@ -252,6 +263,51 @@ impl Storage {
         }
     }
 
+    pub(crate) fn inplace_op1(&mut self, l: &Layout, c: &dyn InplaceOp1) -> Result<()> {
+        match self {
+            Self::Cpu(storage) => c.cpu_fwd(storage, l),
+            Self::Cuda(storage) => c.cuda_fwd(storage, l),
+            Self::Metal(storage) => c.metal_fwd(storage, l),
+        }
+    }
+
+    pub(crate) fn inplace_op2(
+        &mut self,
+        l1: &Layout,
+        t2: &Self,
+        l2: &Layout,
+        c: &dyn InplaceOp2,
+    ) -> Result<()> {
+        self.same_device(t2, c.name())?;
+        match (self, t2) {
+            (Self::Cpu(s1), Self::Cpu(s2)) => c.cpu_fwd(s1, l1, s2, l2),
+            (Self::Cuda(s1), Self::Cuda(s2)) => c.cuda_fwd(s1, l1, s2, l2),
+            (Self::Metal(s1), Self::Metal(s2)) => c.metal_fwd(s1, l1, s2, l2),
+            _ => unreachable!(),
+        }
+    }
+
+    pub(crate) fn inplace_op3(
+        &mut self,
+        l1: &Layout,
+        t2: &Self,
+        l2: &Layout,
+        t3: &Self,
+        l3: &Layout,
+        c: &dyn InplaceOp3,
+    ) -> Result<()> {
+        self.same_device(t2, c.name())?;
+        self.same_device(t3, c.name())?;
+        match (self, t2, t3) {
+            (Self::Cpu(s1), Self::Cpu(s2), Self::Cpu(s3)) => c.cpu_fwd(s1, l1, s2, l2, s3, l3),
+            (Self::Cuda(s1), Self::Cuda(s2), Self::Cuda(s3)) => c.cuda_fwd(s1, l1, s2, l2, s3, l3),
+            (Self::Metal(s1), Self::Metal(s2), Self::Metal(s3)) => {
+                c.metal_fwd(s1, l1, s2, l2, s3, l3)
+            }
+            _ => unreachable!(),
+        }
+    }
+
     pub(crate) fn unary_impl<B: op::UnaryOpT>(&self, layout: &Layout) -> Result<Self> {
         match self {
             Storage::Cpu(storage) => {

candle-core/src/tensor.rs

@@ -1,9 +1,7 @@
 //! Tensors are N-dimensional matrixes of elements using a single data type.
 #![allow(clippy::redundant_closure_call)]
 use crate::backend::{BackendDevice, BackendStorage};
-use crate::op::{
-    BackpropOp, BinaryOp, CmpOp, CustomOp1, CustomOp2, CustomOp3, Op, ReduceOp, UnaryOp,
-};
+use crate::op::{BackpropOp, BinaryOp, CmpOp, Op, ReduceOp, UnaryOp};
 use crate::scalar::TensorOrScalar;
 use crate::shape::{Dim, Dims};
 use crate::{bail, storage::Storage, DType, Device, Error, Layout, Result, Shape};
@@ -81,6 +79,9 @@ macro_rules! unary_op {
     ($fn_name:ident, $op_name:ident) => {
         pub fn $fn_name(&self) -> Result<Self> {
             let shape = self.shape();
+            if shape.elem_count() == 0 {
+                return Ok(self.clone());
+            }
             let storage = self
                 .storage()
                 .unary_impl::<crate::op::$op_name>(self.layout())?;
@@ -94,6 +95,9 @@ macro_rules! binary_op {
     ($fn_name:ident, $op_name:ident) => {
         pub fn $fn_name(&self, rhs: &Self) -> Result<Self> {
             let shape = self.same_shape_binary_op(rhs, stringify!($fn_name))?;
+            if shape.elem_count() == 0 {
+                return Ok(self.clone());
+            }
             let storage = self.storage().binary_impl::<crate::op::$op_name>(
                 &*rhs.storage(),
                 self.layout(),
@@ -116,6 +120,9 @@ macro_rules! binary_op_scalar {
                     .broadcast_as(self.shape())?,
             };
             let shape = self.same_shape_binary_op(&rhs, stringify!($fn_name))?;
+            if self.elem_count() == 0 {
+                return Ok(self.clone());
+            }
             let storage = self.storage().binary_impl::<crate::op::$op_name>(
                 &*rhs.storage(),
                 self.layout(),
@@ -512,6 +519,7 @@ impl Tensor {
     unary_op!(ceil, Ceil);
     unary_op!(floor, Floor);
     unary_op!(round, Round);
+    unary_op!(sign, Sign);
 
     /// Round element of the input tensor to the nearest integer.
     ///
@@ -647,6 +655,9 @@ impl Tensor {
     /// # Ok::<(), candle_core::Error>(())
     /// ```
     pub fn affine(&self, mul: f64, add: f64) -> Result<Self> {
+        if self.elem_count() == 0 {
+            return Ok(self.clone());
+        }
         let storage = self.storage().affine(self.layout(), mul, add)?;
         let op = BackpropOp::new1(self, |arg| Op::Affine { arg, mul, add });
         Ok(from_storage(storage, self.shape(), op, false))
@@ -654,6 +665,9 @@ impl Tensor {
 
     /// Applies the Exponential Linear Unit (ELU) function on each element of the input tensor.
     pub fn elu(&self, alpha: f64) -> Result<Self> {
+        if self.elem_count() == 0 {
+            return Ok(self.clone());
+        }
         let storage = self.storage().elu(self.layout(), alpha)?;
         let op = BackpropOp::new1(self, |t| Op::Elu(t, alpha));
         Ok(from_storage(storage, self.shape(), op, false))
@@ -661,6 +675,9 @@ impl Tensor {
 
     /// Raise the tensor to some float exponent `e`.
     pub fn powf(&self, e: f64) -> Result<Self> {
+        if self.elem_count() == 0 {
+            return Ok(self.clone());
+        }
         let storage = self.storage().powf(self.layout(), e)?;
         let op = BackpropOp::new1(self, |t| Op::Powf(t, e));
         Ok(from_storage(storage, self.shape(), op, false))
@@ -1155,6 +1172,9 @@ impl Tensor {
         let n = b_dims[dim - 1];
 
         let c_shape = Shape::from(&a_dims[..dim - 2]).extend(&[m, n]);
+        if c_shape.elem_count() == 0 || k == 0 {
+            return Tensor::zeros(c_shape, self.dtype(), self.device());
+        }
         let batching: usize = a_dims[..dim - 2].iter().product();
         let batching_b: usize = b_dims[..dim - 2].iter().product();
         if k != k2 || batching != batching_b {
@@ -1351,7 +1371,7 @@ impl Tensor {
             }
             .bt())?
         }
-        let mut storage = self.device().zeros(self.shape(), self.dtype())?;
+        let mut storage = unsafe { self.device().alloc_uninit(self.shape(), self.dtype())? };
         self.storage()
             .copy_strided_src(&mut storage, 0, self.layout())?;
         let offset = start * src.dims()[1..].iter().product::<usize>();
@@ -2001,7 +2021,7 @@ impl Tensor {
             Ok(self.clone())
         } else {
             let shape = self.shape();
-            let mut storage = self.device().zeros(shape, self.dtype())?;
+            let mut storage = unsafe { self.device().alloc_uninit(shape, self.dtype())? };
             self.storage()
                 .copy_strided_src(&mut storage, 0, self.layout())?;
             let op = BackpropOp::new1(self, Op::Copy);
@@ -2009,11 +2029,21 @@ impl Tensor {
         }
     }
 
+    /// Returns a tensor that is in row major order. This always makes a copy.
+    pub fn force_contiguous(&self) -> Result<Tensor> {
+        let shape = self.shape();
+        let mut storage = unsafe { self.device().alloc_uninit(shape, self.dtype())? };
+        self.storage()
+            .copy_strided_src(&mut storage, 0, self.layout())?;
+        let op = BackpropOp::new1(self, Op::Copy);
+        Ok(from_storage(storage, shape.clone(), op, false))
+    }
+
     /// Create a variable based on the values currently stored in a tensor. The storage is always
     /// copied.
     pub(crate) fn make_var(&self) -> Result<Tensor> {
         let shape = self.shape().clone();
-        let mut storage = self.device().zeros(&shape, self.dtype())?;
+        let mut storage = unsafe { self.device().alloc_uninit(&shape, self.dtype())? };
         self.storage()
             .copy_strided_src(&mut storage, 0, self.layout())?;
         Ok(from_storage(storage, shape, BackpropOp::none(), true))
@@ -2066,7 +2096,7 @@ impl Tensor {
             };
             Ok(Tensor(Arc::new(tensor_)))
         } else {
-            let mut storage = self.device().zeros(&shape, self.dtype())?;
+            let mut storage = unsafe { self.device().alloc_uninit(&shape, self.dtype())? };
             self.storage()
                 .copy_strided_src(&mut storage, 0, self.layout())?;
             Ok(from_storage(storage, shape, op, false))
@@ -2093,8 +2123,19 @@ impl Tensor {
         let dim = dim.to_index(self.shape(), "squeeze")?;
         if dims[dim] == 1 {
             let mut dims = dims.to_vec();
+            let mut strides = self.stride().to_vec();
             dims.remove(dim);
-            self.reshape(dims)
+            strides.remove(dim);
+            let tensor_ = Tensor_ {
+                id: TensorId::new(),
+                storage: self.storage.clone(),
+                layout: Layout::new(dims.into(), strides, self.layout.start_offset()),
+                op: BackpropOp::new1(self, Op::Reshape),
+                is_variable: false,
+                dtype: self.dtype,
+                device: self.device.clone(),
+            };
+            Ok(Tensor(Arc::new(tensor_)))
         } else {
             Ok(self.clone())
         }
@@ -2115,10 +2156,24 @@ impl Tensor {
     /// ```
     pub fn unsqueeze<D: Dim>(&self, dim: D) -> Result<Self> {
         let mut dims = self.dims().to_vec();
+        let mut strides = self.stride().to_vec();
         let dim = dim.to_index_plus_one(self.shape(), "unsqueeze")?;
         // Cannot panic because to_index_plus_one already checks dimensions
         dims.insert(dim, 1);
-        self.reshape(dims)
+        // Any stride would work here, but we pick one so as to maximize the probability to remain
+        // C contiguous.
+        let stride = if dim < strides.len() { strides[dim] } else { 1 };
+        strides.insert(dim, stride);
+        let tensor_ = Tensor_ {
+            id: TensorId::new(),
+            storage: self.storage.clone(),
+            layout: Layout::new(dims.into(), strides, self.layout.start_offset()),
+            op: BackpropOp::new1(self, Op::Reshape),
+            is_variable: false,
+            dtype: self.dtype,
+            device: self.device.clone(),
+        };
+        Ok(Tensor(Arc::new(tensor_)))
     }
 
     /// Stacks two or more tensors along a particular dimension.
@@ -2231,6 +2286,10 @@ impl Tensor {
         self.storage.read().unwrap()
     }
 
+    pub(crate) fn storage_mut(&self) -> std::sync::RwLockWriteGuard<'_, Storage> {
+        self.storage.write().unwrap()
+    }
+
     // If we extend the visibility of this function to be usable outside of this crate, we should
     // make it unsafe.
     pub(crate) fn storage_mut_and_layout(
@@ -2252,96 +2311,6 @@ impl Tensor {
         std::ptr::eq(lhs, rhs)
     }
 
-    /// Applies a unary custom op without backward support
-    pub fn apply_op1_no_bwd<C: CustomOp1>(&self, c: &C) -> Result<Self> {
-        let (storage, shape) = self.storage().apply_op1(self.layout(), c)?;
-        Ok(from_storage(storage, shape, BackpropOp::none(), false))
-    }
-
-    /// Applies a binary custom op without backward support
-    pub fn apply_op2_no_bwd<C: CustomOp2>(&self, rhs: &Self, c: &C) -> Result<Self> {
-        let (storage, shape) =
-            self.storage()
-                .apply_op2(self.layout(), &rhs.storage(), rhs.layout(), c)?;
-        Ok(from_storage(storage, shape, BackpropOp::none(), false))
-    }
-
-    /// Applies a ternary custom op without backward support
-    pub fn apply_op3_no_bwd<C: CustomOp3>(&self, t2: &Self, t3: &Self, c: &C) -> Result<Self> {
-        let (storage, shape) = self.storage().apply_op3(
-            self.layout(),
-            &t2.storage(),
-            t2.layout(),
-            &t3.storage(),
-            t3.layout(),
-            c,
-        )?;
-        Ok(from_storage(storage, shape, BackpropOp::none(), false))
-    }
-
-    /// Applies a unary custom op.
-    pub fn apply_op1_arc(&self, c: Arc<Box<dyn CustomOp1 + Send + Sync>>) -> Result<Self> {
-        let (storage, shape) = self
-            .storage()
-            .apply_op1(self.layout(), c.as_ref().as_ref())?;
-        let op = BackpropOp::new1(self, |s| Op::CustomOp1(s, c.clone()));
-        Ok(from_storage(storage, shape, op, false))
-    }
-
-    pub fn apply_op1<C: 'static + CustomOp1 + Send + Sync>(&self, c: C) -> Result<Self> {
-        self.apply_op1_arc(Arc::new(Box::new(c)))
-    }
-
-    /// Applies a binary custom op.
-    pub fn apply_op2_arc(
-        &self,
-        rhs: &Self,
-        c: Arc<Box<dyn CustomOp2 + Send + Sync>>,
-    ) -> Result<Self> {
-        let (storage, shape) = self.storage().apply_op2(
-            self.layout(),
-            &rhs.storage(),
-            rhs.layout(),
-            c.as_ref().as_ref(),
-        )?;
-        let op = BackpropOp::new2(self, rhs, |t1, t2| Op::CustomOp2(t1, t2, c.clone()));
-        Ok(from_storage(storage, shape, op, false))
-    }
-
-    pub fn apply_op2<C: 'static + CustomOp2 + Send + Sync>(&self, r: &Self, c: C) -> Result<Self> {
-        self.apply_op2_arc(r, Arc::new(Box::new(c)))
-    }
-
-    /// Applies a ternary custom op.
-    pub fn apply_op3_arc(
-        &self,
-        t2: &Self,
-        t3: &Self,
-        c: Arc<Box<dyn CustomOp3 + Send + Sync>>,
-    ) -> Result<Self> {
-        let (storage, shape) = self.storage().apply_op3(
-            self.layout(),
-            &t2.storage(),
-            t2.layout(),
-            &t3.storage(),
-            t3.layout(),
-            c.as_ref().as_ref(),
-        )?;
-        let op = BackpropOp::new3(self, t2, t3, |t1, t2, t3| {
-            Op::CustomOp3(t1, t2, t3, c.clone())
-        });
-        Ok(from_storage(storage, shape, op, false))
-    }
-
-    pub fn apply_op3<C: 'static + CustomOp3 + Send + Sync>(
-        &self,
-        t2: &Self,
-        t3: &Self,
-        c: C,
-    ) -> Result<Self> {
-        self.apply_op3_arc(t2, t3, Arc::new(Box::new(c)))
-    }
-
     /// Normalize a 'relative' axis value: positive values are kept, negative
     /// values means counting the dimensions from the back.
     pub fn normalize_axis(&self, axis: i64) -> Result<usize> {

candle-core/src/tensor_cat.rs

@@ -58,20 +58,18 @@ impl Tensor {
                 }
             }
         }
-        if dim == 0 {
+        let all_contiguous = args.iter().all(|v| v.as_ref().is_contiguous());
+        if all_contiguous {
+            Self::cat_contiguous(args, dim)
+        } else if dim == 0 {
             Self::cat0(args)
         } else {
-            let all_contiguous = args.iter().all(|v| v.as_ref().is_contiguous());
-            if all_contiguous {
-                Self::cat_contiguous(args, dim)
-            } else {
-                let args: Vec<Tensor> = args
-                    .iter()
-                    .map(|a| a.as_ref().transpose(0, dim))
-                    .collect::<Result<Vec<_>>>()?;
-                let cat = Self::cat0(&args)?;
-                cat.transpose(0, dim)
-            }
+            let args: Vec<Tensor> = args
+                .iter()
+                .map(|a| a.as_ref().transpose(0, dim))
+                .collect::<Result<Vec<_>>>()?;
+            let cat = Self::cat0(&args)?;
+            cat.transpose(0, dim)
         }
     }
 
@@ -141,7 +139,7 @@ impl Tensor {
         }
         let shape = Shape::from(cat_dims);
         let op = crate::op::BackpropOp::new(args, |args| crate::op::Op::Cat(args, 0));
-        let mut storage = device.zeros(&shape, dtype)?;
+        let mut storage = unsafe { device.alloc_uninit(&shape, dtype)? };
         for (arg, &offset) in args.iter().zip(offsets.iter()) {
             let arg = arg.as_ref();
             arg.storage()
@@ -215,7 +213,7 @@ impl Tensor {
         let block_size: usize = cat_dims.iter().skip(1 + dim).product();
         let shape = Shape::from(cat_dims);
         let op = crate::op::BackpropOp::new(args, |args| crate::op::Op::Cat(args, dim));
-        let mut storage = device.zeros(&shape, dtype)?;
+        let mut storage = unsafe { device.alloc_uninit(&shape, dtype)? };
         let mut dst_o = 0;
         for arg in args.iter() {
             let arg = arg.as_ref();

candle-core/tests/conv_tests.rs

@@ -54,11 +54,6 @@ fn conv1d(dev: &Device) -> Result<()> {
         [2.4509, 2.6357, -1.3336, 4.1393, 0.5657, 1.8091, -1.1784, 3.5675, 0.5069, 3.3352]
     );
 
-    // conv-transposes are not implemented for metal.
-    if dev.is_metal() {
-        return Ok(());
-    }
-
     let w = w.transpose(0, 1)?;
     // The CPU kernels applied in the contiguous and non contiguous cases are different.
     for w in [w.clone(), w.contiguous()?] {
@@ -140,7 +135,7 @@ fn conv2d(dev: &Device) -> Result<()> {
             0.6466, -0.5042, -0.0603, -1.6538, -1.2429, 1.8357, 1.6052, -1.3844, 0.3323, -1.3712,
             0.9634, -0.4799, -0.6451, -0.0840, -1.4247, 0.5512, -0.1747, -0.5509, -0.3742, 0.3790,
             -0.4431, -0.4720, -0.7890, 0.2620, 0.7875, 0.5377, -0.6779, -0.8088, 1.9098, 1.2006,
-            -0.8000, -0.4983, 1.5480, 0.8265, -0.1025, 0.5138, 0.5748, 0.3821, -0.4607, 0.0085,
+            -0.8, -0.4983, 1.5480, 0.8265, -0.1025, 0.5138, 0.5748, 0.3821, -0.4607, 0.0085,
         ],
         dev,
     )?;
@@ -168,33 +163,34 @@ fn conv2d(dev: &Device) -> Result<()> {
             10.389, 3.6023, -4.2808, 0.2672, 5.3646, -5.2023, -2.1955, -9.4075
         ]
     );
-    if !dev.is_metal() {
-        let res = t.conv_transpose2d(&w.transpose(0, 1)?, 0, 0, 1, 1)?;
-        assert_eq!(res.dims(), [1, 2, 7, 7]);
-        assert_eq!(
-            test_utils::to_vec3_round(&res.i(0)?, 4)?,
+
+    let res = t.conv_transpose2d(&w.transpose(0, 1)?, 0, 0, 1, 1)?;
+
+    assert_eq!(res.dims(), [1, 2, 7, 7]);
+    assert_eq!(
+        test_utils::to_vec3_round(&res.i(0)?, 4)?,
+        [
             [
-                [
-                    [-1.9918, 2.6797, -0.4599, -1.6037, 1.4131, -2.4012, 2.9277],
-                    [1.8016, -3.5361, 1.0757, 3.5395, -8.2168, -3.2023, 0.5375],
-                    [0.8243, 1.8675, 7.8929, -4.0746, -6.4415, 5.1139, 1.6889],
-                    [0.2722, 8.9679, 3.3477, 1.8514, -4.2896, -3.8228, -7.5632],
-                    [-8.5412, -5.8142, -7.1587, -1.6095, 0.4651, 0.2748, -2.0985],
-                    [2.0833, -0.6482, -12.1692, -4.1284, -2.9765, -0.0656, -4.5114],
-                    [5.307, 2.6957, 2.3087, 1.0478, 0.7808, -1.1519, -0.9579]
-                ],
-                [
-                    [1.089, 0.1872, -0.6408, -0.9897, 0.8503, 1.1019, -0.9211],
-                    [-0.1741, -0.2915, 4.2472, 1.9417, 1.65, 0.6303, -4.7131],
-                    [1.6555, 2.4026, -2.9293, 2.9953, 0.5328, 3.5873, -0.9621],
-                    [-1.4289, -3.2787, 4.1747, -6.0341, -4.6341, -5.7945, 4.142],
-                    [7.5973, 6.4431, 5.9872, 2.1639, -8.6566, 3.3143, -3.4059],
-                    [-0.8775, -3.048, 11.6543, 0.6442, 2.3218, -0.4765, 1.1516],
-                    [-5.5423, -2.5188, 1.0754, -0.0563, -2.9386, -1.1504, 1.0171]
-                ]
+                [-1.9918, 2.6797, -0.4599, -1.6037, 1.4131, -2.4012, 2.9277],
+                [1.8016, -3.5361, 1.0757, 3.5395, -8.2168, -3.2023, 0.5375],
+                [0.8243, 1.8675, 7.8929, -4.0746, -6.4415, 5.1139, 1.6889],
+                [0.2722, 8.9679, 3.3477, 1.8514, -4.2896, -3.8228, -7.5632],
+                [-8.5412, -5.8142, -7.1587, -1.6095, 0.4651, 0.2748, -2.0985],
+                [2.0833, -0.6482, -12.1692, -4.1284, -2.9765, -0.0656, -4.5114],
+                [5.307, 2.6957, 2.3087, 1.0478, 0.7808, -1.1519, -0.9579]
+            ],
+            [
+                [1.089, 0.1872, -0.6408, -0.9897, 0.8503, 1.1019, -0.9211],
+                [-0.1741, -0.2915, 4.2472, 1.9417, 1.65, 0.6303, -4.7131],
+                [1.6555, 2.4026, -2.9293, 2.9953, 0.5328, 3.5873, -0.9621],
+                [-1.4289, -3.2787, 4.1747, -6.0341, -4.6341, -5.7945, 4.142],
+                [7.5973, 6.4431, 5.9872, 2.1639, -8.6566, 3.3143, -3.4059],
+                [-0.8775, -3.048, 11.6543, 0.6442, 2.3218, -0.4765, 1.1516],
+                [-5.5423, -2.5188, 1.0754, -0.0563, -2.9386, -1.1504, 1.0171]
             ]
-        );
-    }
+        ]
+    );
+
     // Dilations.
     let res = t.conv2d(&w, 0, 1, 2, 1)?;
     assert_eq!(res.dims(), [1, 2, 1, 1]);
@@ -203,44 +199,37 @@ fn conv2d(dev: &Device) -> Result<()> {
         [2.45, -2.3504],
     );
 
-    if !dev.is_metal() {
-        // Transpose and dilations.
-        let res = t.conv_transpose2d(&w.transpose(0, 1)?, 0, 0, 1, 2)?;
-        assert_eq!(res.dims(), [1, 2, 9, 9]);
-        assert_eq!(
-            test_utils::to_vec3_round(&res.i(0)?, 4)?,
+    // Transpose and dilations.
+    let res = t.conv_transpose2d(&w.transpose(0, 1)?, 0, 0, 1, 2)?;
+    assert_eq!(res.dims(), [1, 2, 9, 9]);
+    assert_eq!(
+        test_utils::to_vec3_round(&res.i(0)?, 4)?,
+        [
             [
-                [
-                    [-1.9918, 3.1652, -0.6778, -4.3442, 4.4351, 0.6652, -3.0124, -0.6031, 2.9277],
-                    [2.7036, -1.7156, -0.3969, 1.0516, 1.6381, -2.8886, -0.205, 2.4682, -1.0499],
-                    [-0.9459, 3.1631, 3.707, -4.8369, -8.5166, -1.4496, -2.7559, -3.2698, 1.4376],
-                    [-0.2157, 3.7786, -2.0252, -4.2633, 3.6731, -1.5142, 5.9391, -0.2622, -0.141],
-                    [-6.8121, -3.1744, 1.5945, 3.0637, -9.6088, 1.4446, 2.9489, -3.0082, -7.3822],
-                    [0.2371, 3.3303, 0.3861, 2.2646, -4.6784, 4.1235, -0.0109, 0.3176, -0.03],
-                    [
-                        -2.5339, -2.9564, -3.4518, -4.4594, -9.1873, -1.9709, -0.4676, 0.51,
-                        -3.5024
-                    ],
-                    [4.007, 0.3067, -2.2954, 1.1105, -0.1992, 1.6372, -2.9268, 0.2807, -1.2787],
-                    [5.307, 1.1317, 1.3518, 0.9049, 3.8116, -0.4075, -0.8874, -0.2241, -0.9579]
-                ],
-                [
-                    [1.089, -0.6483, 0.0726, -0.4752, -1.3283, 1.7103, 1.0703, 0.1076, -0.9211],
-                    [-0.8629, 0.1376, 0.3202, 2.0955, 0.9696, 2.8988, -1.0012, 1.5049, -0.1278],
-                    [1.9286, -1.5255, -2.9563, 2.4589, 3.3611, -0.6951, 0.3525, -1.7724, -5.9861],
-                    [1.1226, 2.1561, 3.6417, 4.7546, -0.692, 4.4126, -5.1902, 6.0805, 2.3185],
-                    [1.0111, 0.3604, 0.6432, -3.6605, 7.9517, -9.2955, -5.2988, -3.7803, -2.0642],
-                    [3.3172, -1.7967, -3.6576, -2.0942, 1.3158, 0.112, -1.7405, 2.9167, 0.7957],
-                    [5.1001, 1.8995, -1.8639, 1.1262, 9.9629, 2.683, -3.6319, -1.1607, 0.5856],
-                    [-4.8445, -0.5642, 4.2317, 0.0856, 1.2267, -0.5712, 1.736, 1.0997, 0.6908],
-                    [
-                        -5.5423, -1.1831, -1.2176, 0.0843, 0.0446, -0.7545, -2.4798, -0.0827,
-                        1.0171
-                    ]
-                ]
+                [-1.9918, 3.1652, -0.6778, -4.3442, 4.4351, 0.6652, -3.0124, -0.6031, 2.9277],
+                [2.7036, -1.7156, -0.3969, 1.0516, 1.6381, -2.8886, -0.205, 2.4682, -1.0499],
+                [-0.9459, 3.1631, 3.707, -4.8369, -8.5166, -1.4496, -2.7559, -3.2698, 1.4376],
+                [-0.2157, 3.7786, -2.0252, -4.2633, 3.6731, -1.5142, 5.9391, -0.2622, -0.141],
+                [-6.8121, -3.1744, 1.5945, 3.0637, -9.6088, 1.4446, 2.9489, -3.0082, -7.3822],
+                [0.2371, 3.3303, 0.3861, 2.2646, -4.6784, 4.1235, -0.0109, 0.3176, -0.03],
+                [-2.5339, -2.9564, -3.4518, -4.4594, -9.1873, -1.9709, -0.4676, 0.51, -3.5024],
+                [4.007, 0.3067, -2.2954, 1.1105, -0.1992, 1.6372, -2.9268, 0.2807, -1.2787],
+                [5.307, 1.1317, 1.3518, 0.9049, 3.8116, -0.4075, -0.8874, -0.2241, -0.9579]
+            ],
+            [
+                [1.089, -0.6483, 0.0726, -0.4752, -1.3283, 1.7103, 1.0703, 0.1076, -0.9211],
+                [-0.8629, 0.1376, 0.3202, 2.0955, 0.9696, 2.8988, -1.0012, 1.5049, -0.1278],
+                [1.9286, -1.5255, -2.9563, 2.4589, 3.3611, -0.6951, 0.3525, -1.7724, -5.9861],
+                [1.1226, 2.1561, 3.6417, 4.7546, -0.692, 4.4126, -5.1902, 6.0805, 2.3185],
+                [1.0111, 0.3604, 0.6432, -3.6605, 7.9517, -9.2955, -5.2988, -3.7803, -2.0642],
+                [3.3172, -1.7967, -3.6576, -2.0942, 1.3158, 0.112, -1.7405, 2.9167, 0.7957],
+                [5.1001, 1.8995, -1.8639, 1.1262, 9.9629, 2.683, -3.6319, -1.1607, 0.5856],
+                [-4.8445, -0.5642, 4.2317, 0.0856, 1.2267, -0.5712, 1.736, 1.0997, 0.6908],
+                [-5.5423, -1.1831, -1.2176, 0.0843, 0.0446, -0.7545, -2.4798, -0.0827, 1.0171]
             ]
-        );
-    }
+        ]
+    );
+
     Ok(())
 }
 
@@ -287,19 +276,13 @@ fn conv2d_small(dev: &Device) -> Result<()> {
     assert_eq!(
         test_utils::to_vec1_round(&res.flatten_all()?, 4)?,
         [
-            0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,
-            0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.1640, -0.0111, -0.1742, 0.0000, 0.0000,
-            0.0000, 0.0000, 2.6437, -2.0268, 1.1823, 0.0000, 0.0000, 0.0000, 0.0000, 3.2855,
-            -1.0324, 0.2539, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,
-            0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000
+            0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.1640,
+            -0.0111, -0.1742, 0.0, 0.0, 0.0, 0.0, 2.6437, -2.0268, 1.1823, 0.0, 0.0, 0.0, 0.0,
+            3.2855, -1.0324, 0.2539, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+            0.0, 0.0, 0.0, 0.0
         ]
     );
 
-    // conv-transposes are not implemented for metal
-    if dev.is_metal() {
-        return Ok(());
-    }
-
     let res = t.conv_transpose2d(&w.transpose(0, 1)?, 0, 0, 1, 1)?;
     assert_eq!(res.dims(), [1, 1, 3, 3]);
     assert_eq!(
@@ -402,9 +385,6 @@ print(w.grad[0])
 */
 fn conv2d_grad(dev: &Device) -> Result<()> {
     // conv-transposes are not implemented for metal
-    if dev.is_metal() {
-        return Ok(());
-    }
     use candle_core::Var;
     let t = Var::from_slice(
         &[
@@ -417,7 +397,7 @@ fn conv2d_grad(dev: &Device) -> Result<()> {
             0.6466, -0.5042, -0.0603, -1.6538, -1.2429, 1.8357, 1.6052, -1.3844, 0.3323, -1.3712,
             0.9634, -0.4799, -0.6451, -0.0840, -1.4247, 0.5512, -0.1747, -0.5509, -0.3742, 0.3790,
             -0.4431, -0.4720, -0.7890, 0.2620, 0.7875, 0.5377, -0.6779, -0.8088, 1.9098, 1.2006,
-            -0.8000, -0.4983, 1.5480, 0.8265, -0.1025, 0.5138, 0.5748, 0.3821, -0.4607, 0.0085,
+            -0.8, -0.4983, 1.5480, 0.8265, -0.1025, 0.5138, 0.5748, 0.3821, -0.4607, 0.0085,
         ],
         (1, 4, 5, 5),
         dev,
@@ -602,6 +582,154 @@ fn conv2d_grad(dev: &Device) -> Result<()> {
         ]
     );
 
+    // Conv Transpose 2d Test
+    //tested against following python
+
+    // import torch
+    // torch.manual_seed(4242)
+    // padding = 4
+    // outpadding = 2
+    // dilation = 3
+    // stride = 3
+    // input = torch.randn((1, 4, 7, 5), requires_grad=True)
+    // kernel = torch.randn((4, 2, 3, 5), requires_grad=True)
+    // print("input", input.flatten())
+    // print("kernel", kernel.flatten())
+    // res = torch.nn.functional.conv_transpose2d(
+    //     input,
+    //     kernel,
+    //     stride=stride,
+    //     padding=padding,
+    //     dilation=dilation,
+    //     output_padding=outpadding,
+    // )
+    // res.retain_grad()
+    // print(res.shape)
+    // loss = (res**2).sum()
+    // print(loss)
+    // loss.backward()
+    // print(input.grad.shape)
+    // print("input grad", torch.round(input.grad, decimals=1))
+    // print(kernel.grad.shape)
+    // print("kernel grad", torch.round(kernel.grad.flatten(), decimals=1))
+
+    let padding = 4;
+    let outpadding = 2;
+    let dilation = 3;
+    let stride = 3;
+
+    let t = Var::from_slice(
+        &[
+            0.4056_f32, -0.8689, -0.0773, -1.5630, -2.8012, -1.5059, 0.3972, 1.0852, 0.4997,
+            3.0616, 1.6541, 0.0964, -0.8338, -1.6523, -0.8323, -0.1699, 0.0823, 0.3526, 0.6843,
+            0.2395, 1.2279, -0.9287, -1.7030, 0.1370, 0.6047, 0.3770, -0.6266, 0.3529, 2.2013,
+            -0.6836, 0.2477, 1.3127, -0.2260, 0.2622, -1.2974, -0.8140, -0.8404, -0.3490, 0.0130,
+            1.3123, 1.7569, -0.3956, -1.8255, 0.1727, -0.3538, 2.6941, 1.0529, 0.4219, -0.2071,
+            1.1586, 0.4717, 0.3865, -0.5690, -0.5010, -0.1310, 0.7796, 0.6630, -0.2021, 2.6090,
+            0.2049, 0.6466, -0.5042, -0.0603, -1.6538, -1.2429, 1.8357, 1.6052, -1.3844, 0.3323,
+            -1.3712, 0.9634, -0.4799, -0.6451, -0.0840, -1.4247, 0.5512, -0.1747, -0.5509, -0.3742,
+            0.3790, -0.4431, -0.4720, -0.7890, 0.2620, 0.5411, -1.1715, -2.4997, 2.3249, -0.8912,
+            -0.4733, -0.5701, -2.8888, -1.4112, -0.5471, -0.9234, -1.1660, 0.4189, -0.7465,
+            -0.6473, 0.1402, 0.7875, 0.5377, -0.6779, -0.8088, -0.4864, -0.2312, 0.9279, 0.1264,
+            1.5480, 0.8265, -0.1025, 0.5138, -0.2512, 0.1576, 1.2705, 0.3641, -0.9325, 0.6451,
+            -0.8537, 0.2378, 0.1794, 0.2752, -0.3687, -1.1149, -0.1410, -0.5829, -0.0892, 1.4258,
+            -2.2789, 0.5270, 0.1825, 1.7007, -0.5263, -0.2954, 0.4440, 0.5537, 0.3492, 0.6186,
+            1.6475, 0.2219,
+        ],
+        (1, 4, 7, 5),
+        dev,
+    )?;
+
+    #[rustfmt::skip]
+    let w = Var::from_slice(
+        &[
+            -1.1744_f32, 0.3266, 2.5893, 1.0142, 0.1763, 0.7752, 0.6604, 0.2029, -0.2145, 0.7234,
+            -0.3441, -1.5400, -0.6333, 0.6613, 0.2083, 0.6230, -1.7002, 0.3393, 0.4049, 1.0762,
+            0.2723, 1.4181, 0.0029, -0.2122, 1.7668, 1.4168, 0.3320, -0.2719, 0.7932, -0.7204,
+            0.4447, 0.1211, 0.5908, 1.0089, -0.1646, 1.8033, -0.6286, 0.2016, -0.3370, 1.2555,
+            0.8009, -0.6488, -0.4652, -1.5685, 1.5860, 0.5583, 0.4623, 0.6026, 0.8828, 2.4990,
+            0.6811, -0.3369, 1.3320, 1.7669, -1.1067, 1.2958, -0.9415, -0.9655, -0.4462, 0.7181,
+            0.5181, -1.1658, -1.8467, -0.7763, 1.2769, 0.8651, 0.9890, 1.5092, 0.7207, -0.8481,
+            0.7417, 0.3375, -1.2685, 1.4572, 1.0915, 0.1093, -0.8550, -0.5831, -0.6309, -0.2509,
+            0.5220, -0.0914, 0.7900, 0.1096, 0.3258, 0.2723, -1.0942, -0.3393, -0.1653, 0.5732,
+            -0.8014, 1.8194, -1.9023, 0.2127, 1.8636, -0.8979, 0.1927, -0.2778, 0.3105, 0.0071,
+            -1.1823, 0.2476, -0.7178, -1.3821, 1.0769, -0.4376, -0.9967, -0.1227, 1.6197, -1.0604,
+            0.1372, 0.8141, -0.6163, 0.7304, -0.8285, 2.0636, -0.7176, 0.2495, -0.2581, -0.4478,
+        ],
+        (4, 2, 3, 5),
+        dev,
+    )?;
+    let res = t.conv_transpose2d(&w, padding, outpadding, stride, dilation)?;
+    let loss = res.sqr()?.sum_all()?;
+    assert_eq!(test_utils::to_vec0_round(&loss, 0)?, 2904.0);
+    let grads = loss.backward()?;
+
+    let grad_t = grads.get(&t).unwrap();
+    let grad_w = grads.get(&w).unwrap();
+    assert_eq!(grad_t.dims(), [1, 4, 7, 5]);
+    assert_eq!(grad_w.dims(), [4, 2, 3, 5]);
+
+    assert_eq!(
+        test_utils::to_vec1_round(&grad_w.flatten_all()?, 1)?,
+        [
+            // torch gets 89.1
+            -89.0, -135.3, 136.7, 102.0, -53.4, 117.9, 118.6, -43.9, -218.0, -58.5, -114.3, -150.0,
+            -15.6, 172.1, 66.3, -64.3, -27.9, -19.8, 31.7, 62.1, 5.5, 92.6, 28.2, -29.6, 55.9,
+            52.7, -72.7, -119.8, 53.8, -25.5, 128.8, 19.3, 68.0, 190.9, -64.1, -86.2, -111.2,
+            106.6, -67.7, 37.8, 115.9, 50.4, -77.7, -54.9, 22.3, -4.6, 89.8, 61.7, 122.4, 192.6,
+            -27.8, -104.6, 57.0, 166.4, 27.1, 6.1, 18.7, -93.2, 31.5, 168.2, -3.7, -99.5, -55.5,
+            -10.8, 17.5, 20.8, 16.9, 43.8, 42.0, -89.2, 18.8, -9.6, -84.1, 212.6, 19.7, -50.0,
+            -52.0, -40.0, -166.6, -73.2, -10.8, -73.3, 31.5, -23.4, -79.3, -27.0, -84.4, -42.9,
+            -20.3, 51.8, -16.7, 76.3, -120.5, -65.8, 96.5, -10.7, -45.9, -88.1, 65.4, -7.0, -1.5,
+            92.8, -25.1, -114.2, -5.8, -14.8, -51.2, -20.7, 54.2, -79.8, 47.7, -29.2, -8.8, 53.5,
+            -28.4, 85.0, -18.3, 107.0, 28.3, -71.8
+        ]
+    );
+
+    assert_eq!(
+        test_utils::to_vec3_round(&grad_t.i(0)?, 1)?,
+        [
+            [
+                [32.3, -41.6, -24.0, 14.1, 17.6],
+                [-11.8, 72.5, 87.6, 46.4, 61.5],
+                [115.0, 108.5, -48.6, -63.4, -50.0],
+                [51.3, 5.4, 31.3, 91.1, -30.9],
+                [52.7, 92.8, -68.0, -47.0, 83.0],
+                // pytorch gets -107.1
+                [-10.2, -107.0, -5.4, 213.1, -31.4],
+                [-2.4, 65.1, 9.2, -146.2, -24.2]
+            ],
+            [
+                [-72.6, -63.9, -61.9, 45.3, 33.0],
+                [79.3, -0.5, -26.2, 78.2, 42.7],
+                [90.9, 141.6, 40.1, -62.7, 37.0],
+                [32.8, 198.2, -0.8, -31.1, 27.3],
+                // torch gets 48.0
+                [34.5, 34.9, -47.9, 127.6, -12.3],
+                [-61.4, -3.2, -2.9, -10.9, -16.6],
+                [74.6, 60.1, -68.9, 34.5, -50.4]
+            ],
+            [
+                [37.5, -56.9, -43.6, -13.5, -9.9],
+                [40.0, 97.3, 28.6, 14.2, -30.1],
+                [-22.3, -126.3, -68.8, -8.2, 26.1],
+                [-32.9, 37.3, 108.5, -54.8, 29.6],
+                [34.9, -176.9, -125.0, -28.3, -13.9],
+                [-54.9, 142.6, 62.1, -80.4, -65.6],
+                [7.4, -91.1, -67.6, 35.0, 39.7]
+            ],
+            [
+                [-57.2, -40.9, -10.1, 32.6, 29.4],
+                [18.7, -18.0, 29.5, -1.2, 59.2],
+                [-14.0, -74.4, 19.8, -117.0, 58.2],
+                [-21.8, 163.5, -71.1, -99.0, 80.9],
+                [-58.9, -10.9, 93.8, -139.6, 98.0],
+                // torch gets 54.5
+                [-54.4, 135.3, 6.0, -79.1, 134.6],
+                [27.5, -76.0, 43.4, -2.8, -7.8]
+            ]
+        ]
+    );
     Ok(())
 }
 

candle-core/tests/custom_op_tests.rs

@@ -112,3 +112,34 @@ fn custom_op1_with_backward() -> Result<()> {
 
     Ok(())
 }
+
+impl candle_core::InplaceOp1 for Elu {
+    fn name(&self) -> &'static str {
+        "elu"
+    }
+
+    fn cpu_fwd(&self, s: &mut CpuStorage, _l: &Layout) -> Result<()> {
+        let alpha = self.alpha;
+        match s {
+            CpuStorage::BF16(s) => s.iter_mut().for_each(|v| *v = fwd(*v, alpha)),
+            CpuStorage::F16(s) => s.iter_mut().for_each(|v| *v = fwd(*v, alpha)),
+            CpuStorage::F32(s) => s.iter_mut().for_each(|v| *v = fwd(*v, alpha)),
+            CpuStorage::F64(s) => s.iter_mut().for_each(|v| *v = fwd(*v, alpha)),
+            _ => candle_core::bail!("unsupported dtype for inplace elu"),
+        }
+        Ok(())
+    }
+}
+
+#[test]
+fn inplace_op1() -> Result<()> {
+    let cpu = &Device::Cpu;
+    let t = Tensor::arange(0u32, 12u32, cpu)?.to_dtype(DType::F32)?;
+    let t = (t - 5.)?;
+    t.inplace_op1(&Elu { alpha: 1. })?;
+    assert_eq!(
+        to_vec1_round(&t, 4)?,
+        &[-0.9933, -0.9817, -0.9502, -0.8647, -0.6321, 0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0]
+    );
+    Ok(())
+}

candle-core/tests/layout_tests.rs

@@ -88,7 +88,7 @@ fn strided_blocks() -> Result<()> {
         }
     };
     let tensor = Tensor::arange(0u32, 24u32, &Cpu)?.reshape((2, 3, 4))?;
-    let tensor = tensor.i((.., 1))?;
+    let tensor = tensor.i((.., 1))?.contiguous()?;
     match tensor.strided_blocks() {
         candle::StridedBlocks::SingleBlock { start_offset, len } => {
             assert_eq!(start_offset, 0);
@@ -100,6 +100,20 @@ fn strided_blocks() -> Result<()> {
         }
     };
     let tensor = Tensor::arange(0u32, 24u32, &Cpu)?.reshape((2, 3, 4))?;
+    let tensor = tensor.i((.., 1))?;
+    match tensor.strided_blocks() {
+        candle::StridedBlocks::SingleBlock { .. } => {
+            panic!("unexpected block structure")
+        }
+        candle::StridedBlocks::MultipleBlocks {
+            block_len,
+            block_start_index,
+        } => {
+            assert_eq!(block_len, 4);
+            assert_eq!(block_start_index.collect::<Vec<_>>(), &[4, 16])
+        }
+    };
+    let tensor = Tensor::arange(0u32, 24u32, &Cpu)?.reshape((2, 3, 4))?;
     match tensor.t()?.strided_blocks() {
         candle::StridedBlocks::SingleBlock { .. } => {
             panic!("unexpected block structure")

candle-core/tests/matmul_tests.rs

@@ -0,0 +1,106 @@
+use candle_core::{test_device, DType, Device, IndexOp, Result, Tensor};
+
+fn matmul(device: &Device) -> Result<()> {
+    let data = vec![1.0f32, 2.0, 3.0, 4.0];
+    let a = Tensor::from_slice(&data, (2, 2), device)?;
+    let data = vec![1.0f32, 2.0, 3.0, 4.0];
+    let b = Tensor::from_slice(&data, (2, 2), device)?;
+
+    let c = a.matmul(&b)?;
+    assert_eq!(c.to_vec2::<f32>()?, &[[7.0f32, 10.0], [15.0, 22.0]]);
+
+    let data = vec![1.0f32, 2.0];
+    let a = Tensor::from_slice(&data, (2, 1), device)?;
+    let data = vec![3.0f32, 4.0];
+    let b = Tensor::from_slice(&data, (1, 2), device)?;
+    let c = a.matmul(&b)?;
+    assert_eq!(c.to_vec2::<f32>()?, &[&[3.0, 4.0], &[6.0, 8.0]]);
+
+    let data: Vec<_> = (0..6).map(|i| i as f32).collect();
+    let a = Tensor::from_slice(&data, (2, 3), device)?;
+    let data: Vec<_> = (0..6).map(|i| (i + 2) as f32).collect();
+    let b = Tensor::from_slice(&data, (3, 2), device)?;
+    let c = a.matmul(&b)?;
+    assert_eq!(c.to_vec2::<f32>()?, &[&[16., 19.], &[52., 64.]]);
+
+    let data: Vec<_> = (0..12).map(|i| i as f32).collect();
+    let a = Tensor::from_slice(&data, (2, 2, 3), device)?;
+    let data: Vec<_> = (0..12).map(|i| (i + 2) as f32).collect();
+    let b = Tensor::from_slice(&data, (2, 3, 2), device)?;
+    let expected = [[[16., 19.], [52., 64.]], [[214., 235.], [304., 334.]]];
+
+    let c = a.matmul(&b)?;
+    assert_eq!(c.to_vec3::<f32>()?, &expected);
+
+    // Also perform the matmul on contiguous transposed versions.
+    let a_tt = a.t()?.contiguous()?.t()?;
+    assert!(!a_tt.is_contiguous());
+    assert_eq!(a.dims(), a_tt.dims());
+    assert_eq!(a_tt.stride(), &[6, 1, 2]);
+
+    let b_tt = b.t()?.contiguous()?.t()?;
+    assert!(!b_tt.is_contiguous());
+    assert_eq!(b.dims(), b_tt.dims());
+    assert_eq!(b_tt.stride(), &[6, 1, 3]);
+
+    assert_eq!(a_tt.matmul(&b)?.to_vec3::<f32>()?, &expected);
+    assert_eq!(a.matmul(&b_tt)?.to_vec3::<f32>()?, &expected);
+    assert_eq!(a_tt.matmul(&b_tt)?.to_vec3::<f32>()?, &expected);
+    Ok(())
+}
+
+fn broadcast_matmul(device: &Device) -> Result<()> {
+    let lhs = Tensor::randn(0f32, 1f32, (3, 1, 4, 5), device)?;
+    let rhs = Tensor::randn(0f32, 1f32, (6, 5, 2), device)?;
+    let out = lhs.broadcast_matmul(&rhs)?;
+    assert_eq!(out.dims(), &[3, 6, 4, 2]);
+    for idx1 in 0..3 {
+        for idx2 in 0..6 {
+            let out = out.i((idx1, idx2))?;
+            let lhs = lhs.i((idx1, 0))?;
+            let rhs = rhs.i(idx2)?;
+            let out2 = lhs.matmul(&rhs);
+            let sum_diff2 = (out - out2)?.sqr()?.sum_all()?;
+            // With cuda, we see errors of up to ~1e-12.
+            assert!(sum_diff2.to_vec0::<f32>()? < 1e-6)
+        }
+    }
+    Ok(())
+}
+
+// https://github.com/huggingface/candle/issues/1948
+fn squeeze_mm(device: &Device) -> Result<()> {
+    let seq_len = 8_usize;
+    let a = Tensor::zeros((1, seq_len, 16), DType::F32, device)?;
+    let x = a.i((.., seq_len - 1, ..))?;
+    let w = Tensor::zeros((32, 16), DType::F32, device)?.t()?;
+    let x = x.matmul(&w)?;
+    assert_eq!(x.dims(), &[1, 32]);
+    Ok(())
+}
+
+// https://github.com/huggingface/candle/issues/1992
+fn mm_layout(device: &Device) -> Result<()> {
+    let a = Tensor::arange(0f32, 16f32, device)?.reshape((1, 1, 4, 4))?;
+    let b = Tensor::arange(0f32, 8f32, device)?.reshape((1, 1, 4, 2))?;
+    let mm1 = a.matmul(&b)?;
+    // Forces the layout to be:
+    // shape: [1, 1, 4, 2], stride: [8, 2, 2, 1], start_offset: 0
+    // This is still a contiguous matrix but matmul checks are only the two last dimensions have
+    // non 1 sizes but matmul check may be reluctant to handle it.
+    let b = b.transpose(1, 2)?.force_contiguous()?.transpose(1, 2)?;
+    let mm2 = a.matmul(&b)?;
+    let diff = (mm1 - mm2)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    assert_eq!(diff, 0.);
+    Ok(())
+}
+
+test_device!(matmul, matmul_cpu, matmul_gpu, matmul_metal);
+test_device!(
+    broadcast_matmul,
+    broadcast_matmul_cpu,
+    broadcast_matmul_gpu,
+    broadcast_matmul_metal
+);
+test_device!(squeeze_mm, squeeze_mm_cpu, squeeze_mm_gpu, squeeze_mm_metal);
+test_device!(mm_layout, mm_layout_cpu, mm_layout_gpu, mm_layout_metal);

candle-core/tests/pool_tests.rs

@@ -2,9 +2,6 @@ use candle_core::{test_device, test_utils, Device, IndexOp, Result, Tensor};
 
 // https://github.com/huggingface/candle/issues/364
 fn avg_pool2d(dev: &Device) -> Result<()> {
-    if dev.is_metal() {
-        return Ok(());
-    }
     let data: Vec<f32> = vec![
         1., 1., 1., 1., 0., 0., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
     ];
@@ -22,9 +19,6 @@ fn avg_pool2d(dev: &Device) -> Result<()> {
 }
 
 fn max_pool2d(dev: &Device) -> Result<()> {
-    if dev.is_metal() {
-        return Ok(());
-    }
     let data: Vec<f32> = vec![
         1., 2., 1., 3., 0., 0., 1., 1., 1., 1., 1., 1., 5., 1., 1., 1.,
     ];

candle-core/tests/quantized_tests.rs

@@ -3,7 +3,7 @@ use candle_core::{
     quantized::{self, GgmlDType},
     test_device,
     test_utils::to_vec2_round,
-    Device, Module, Result, Tensor,
+    Device, IndexOp, Module, Result, Tensor,
 };
 use quantized::{k_quants, GgmlType};
 use rand::prelude::*;
@@ -47,18 +47,14 @@ fn test_matmul(
 }
 
 fn quantized_matmul(device: &Device) -> Result<()> {
-    // TODO Enable this later when we enable cuda.
-    if device.is_cuda() {
-        return Ok(());
-    }
     let (m, k, n) = (3, 64, 4);
-    let lhs = (0..(m * k)).map(|v| v as f32).collect::<Vec<_>>();
-    let tensor_lhs = Tensor::from_slice(&lhs, (m, k), device)?;
+    let lhs_s = (0..(m * k)).map(|v| v as f32).collect::<Vec<_>>();
+    let lhs = Tensor::from_slice(&lhs_s, (m, k), device)?;
     let mut dst = vec![42.; 3 * 4];
     let mut rhs_t = vec![k_quants::BlockQ4_0::zeros(); 8];
     let rhs = (0..(k * n)).map(|v| v as f32).collect::<Vec<_>>();
     k_quants::BlockQ4_0::from_float(&rhs, &mut rhs_t)?;
-    k_quants::matmul((m, k, n), &lhs, &rhs_t, &mut dst)?;
+    k_quants::matmul((m, k, n), &lhs_s, &rhs_t, &mut dst)?;
     assert_eq!(
         dst.iter().map(|x| x.round()).collect::<Vec<_>>(),
         &[
@@ -67,7 +63,7 @@ fn quantized_matmul(device: &Device) -> Result<()> {
         ]
     );
     let tensor_rhs = Tensor::from_slice(&rhs, (n, k), device)?.t()?;
-    let mm = tensor_lhs.matmul(&tensor_rhs)?;
+    let mm = lhs.matmul(&tensor_rhs)?;
     assert_eq!(
         mm.to_vec2::<f32>()?,
         &[
@@ -79,7 +75,7 @@ fn quantized_matmul(device: &Device) -> Result<()> {
 
     let qtensor = quantized::QTensor::quantize(&tensor_rhs.t()?, GgmlDType::Q4_0)?;
     let matmul = quantized::QMatMul::from_qtensor(qtensor)?;
-    let res = matmul.forward(&tensor_lhs)?;
+    let res = matmul.forward(&lhs)?;
     match device {
         Device::Metal(_) => assert_eq!(
             to_vec2_round(&res, 0)?,
@@ -89,7 +85,15 @@ fn quantized_matmul(device: &Device) -> Result<()> {
                 [341970.0, 994574.0, 1656181.0, 2302182.0]
             ]
         ),
-        _ => assert_eq!(
+        Device::Cuda(_) => assert_eq!(
+            to_vec2_round(&res, 0)?,
+            &[
+                [84866.0, 214045.0, 344676.0, 473707.0],
+                [213425.0, 604313.0, 1000431.0, 1387960.0],
+                [342030.0, 994630.0, 1656248.0, 2302250.0]
+            ]
+        ),
+        Device::Cpu => assert_eq!(
             to_vec2_round(&res, 0)?,
             &[
                 [85120.0, 214562.0, 345455.0, 474748.0],
@@ -98,22 +102,16 @@ fn quantized_matmul(device: &Device) -> Result<()> {
             ]
         ),
     }
-
     test_matmul(device, (1, 3, 4, 256), GgmlDType::Q4_0)?;
-
     Ok(())
 }
 
 fn quantized_matmul_neg(device: &Device) -> Result<()> {
-    // TODO Enable this later when we enable cuda.
-    if device.is_cuda() {
-        return Ok(());
-    }
     let (m, k, n) = (3, 64, 4);
-    let lhs = (0..(m * k))
+    let lhs_s = (0..(m * k))
         .map(|v| v as f32 - (m * k) as f32 / 2.0)
         .collect::<Vec<_>>();
-    let tensor_lhs = Tensor::from_slice(&lhs, (m, k), device)?;
+    let lhs = Tensor::from_slice(&lhs_s, (m, k), device)?;
     let mut dst = vec![42.; 3 * 4];
     let mut rhs_t = vec![k_quants::BlockQ4_0::zeros(); 8];
     let rhs = (0..k * n)
@@ -121,7 +119,7 @@ fn quantized_matmul_neg(device: &Device) -> Result<()> {
         .collect::<Vec<_>>();
     let tensor_rhs = Tensor::from_slice(&rhs, (n, k), device)?.t()?;
     k_quants::BlockQ4_0::from_float(&rhs, &mut rhs_t)?;
-    k_quants::matmul((m, k, n), &lhs, &rhs_t, &mut dst)?;
+    k_quants::matmul((m, k, n), &lhs_s, &rhs_t, &mut dst)?;
     assert_eq!(
         dst.iter().map(|x| x.round()).collect::<Vec<_>>(),
         &[
@@ -129,7 +127,7 @@ fn quantized_matmul_neg(device: &Device) -> Result<()> {
             -196472.0, 63012.0, 324585.0, 587902.0
         ]
     );
-    let mm = tensor_lhs.matmul(&tensor_rhs)?;
+    let mm = lhs.matmul(&tensor_rhs)?;
     assert_eq!(
         to_vec2_round(&mm, 0)?,
         &[
@@ -141,7 +139,7 @@ fn quantized_matmul_neg(device: &Device) -> Result<()> {
 
     let qtensor = quantized::QTensor::quantize(&tensor_rhs.t()?, GgmlDType::Q4_0)?;
     let matmul = quantized::QMatMul::from_qtensor(qtensor)?;
-    let res = matmul.forward(&tensor_lhs)?;
+    let res = matmul.forward(&lhs)?;
     match device {
         Device::Metal(_) => assert_eq!(
             to_vec2_round(&res, 0)?,
@@ -151,7 +149,15 @@ fn quantized_matmul_neg(device: &Device) -> Result<()> {
                 [-196102.0, 63022.0, 324233.0, 587191.0]
             ]
         ),
-        _ => assert_eq!(
+        Device::Cuda(_) => assert_eq!(
+            to_vec2_round(&res, 0)?,
+            &[
+                [243740.0, -19762.0, -285476.0, -550498.0],
+                [23774.0, 21645.0, 19395.0, 18364.0],
+                [-196045.0, 63030.0, 324120.0, 587079.0]
+            ]
+        ),
+        Device::Cpu => assert_eq!(
             to_vec2_round(&res, 0)?,
             &[
                 [243524.0, -19596.0, -285051.0, -549815.0],
@@ -160,22 +166,58 @@ fn quantized_matmul_neg(device: &Device) -> Result<()> {
             ]
         ),
     }
-
+    let lhs2 = Tensor::stack(&[&lhs, &lhs], 0)?;
+    let res2 = matmul.forward(&lhs2)?;
+    let res2 = res2.i(1)?;
+    let diff = (res - res2)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    if device.is_cuda() {
+        assert!(diff < 0.1);
+    } else {
+        assert_eq!(diff, 0.);
+    }
     Ok(())
 }
 
-test_device!(
-    quantized_matmul,
-    quantized_matmul_cpu,
-    quantized_matmul_cuda,
-    quantized_matmul_metal
-);
-test_device!(
-    quantized_matmul_neg,
-    quantized_matmul_neg_cpu,
-    quantized_matmul_neg_cuda,
-    quantized_matmul_neg_metal
-);
+fn qmm_batch(dev: &Device) -> Result<()> {
+    let (lhs, rhs, _mm) = get_random_tensors(2, 256, 6, dev)?;
+    let rhs = quantized::QTensor::quantize(&rhs, GgmlDType::Q2K)?;
+    let rhs = quantized::QMatMul::from_qtensor(rhs)?;
+    let mm = rhs.forward(&lhs)?;
+    assert_eq!(mm.shape().dims(), [2, 6]);
+    let lhs2 = Tensor::cat(&[&lhs, &lhs], 0)?;
+    let mm2 = rhs.forward(&lhs2)?;
+    assert_eq!(mm2.shape().dims(), [4, 6]);
+    let diff2 = (mm2.i(2..)? - &mm)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    assert_eq!(diff2, 0.0);
+    let lhs3 = Tensor::cat(&[&lhs2, &lhs], 0)?;
+    let mm3 = rhs.forward(&lhs3)?;
+    assert_eq!(mm3.shape().dims(), [6, 6]);
+    let diff3 = (mm3.i(2..4)? - &mm)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    assert_eq!(diff3, 0.0);
+    let diff3 = (mm3.i(4..)? - &mm)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    assert_eq!(diff3, 0.0);
+    let lhs4 = Tensor::cat(&[&lhs3, &lhs3], 0)?;
+    let mm4 = rhs.forward(&lhs4)?;
+    assert_eq!(mm4.shape().dims(), [12, 6]);
+    let diff4 = (mm4.i(..6)? - &mm3)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    if dev.is_cuda() {
+        // We use a different kernel for sizes from 1 to 8 on cuda which explains
+        // the difference here.
+        assert!(0. < diff4 && diff4 < 1e-4)
+    } else {
+        assert_eq!(diff4, 0.0)
+    };
+    let diff4 = (mm4.i(6..)? - &mm4.i(..6)?)?
+        .abs()?
+        .sum_all()?
+        .to_vec0::<f32>()?;
+    assert_eq!(diff4, 0.0);
+    Ok(())
+}
+
+test_device!(quantized_matmul, qmm_cpu, qmm_cuda, qmm_metal);
+test_device!(quantized_matmul_neg, qmm_n_cpu, qmm_n_cuda, qmm_n_metal);
+test_device!(qmm_batch, qmm_b_cpu, qmm_b_cuda, qmm_b_metal);
 
 fn quantize_q4_0(device: &Device) -> Result<()> {
     let src = (0..32 * 4).map(|v| v as f32).collect::<Vec<_>>();

candle-core/tests/tensor_tests.rs

@@ -106,6 +106,9 @@ fn unary_op(device: &Device) -> Result<()> {
             [2.6911, -0.0647, -0.1091, 1.7353, 2.7933]
         ]
     );
+    let t_f16 = tensor.to_dtype(DType::F16)?.gelu()?.to_dtype(DType::F32)?;
+    let max_diff = (tensor.gelu()? - t_f16)?.flatten_all()?.max(0)?;
+    assert!(max_diff.to_vec0::<f32>()? < 5e-3);
     assert_eq!(
         test_utils::to_vec2_round(&tensor.gelu_erf()?, 4)?,
         [
@@ -148,6 +151,14 @@ fn unary_op(device: &Device) -> Result<()> {
         test_utils::to_vec1_round(&tensor.round_to(-2)?, 4)?,
         [3000.0, 300.]
     );
+    let tensor = Tensor::new(
+        &[-1.01f32, -0.9, -0.1, 0.0, -0.0, 0.1, 0.9, 1.0, 1.1],
+        device,
+    )?;
+    assert_eq!(
+        tensor.sign()?.to_vec1::<f32>()?,
+        [-1., -1., -1., 0., 0., 1., 1., 1., 1.]
+    );
     Ok(())
 }
 
@@ -707,6 +718,8 @@ fn embeddings(device: &Device) -> Result<()> {
     assert_eq!(hs.to_vec2::<f32>()?, &[[0.0, 1.0], [4.0, 5.0], [2.0, 3.0]]);
     let hs = t.index_select(&ids, 0)?;
     assert_eq!(hs.to_vec2::<f32>()?, &[[0.0, 1.0], [4.0, 5.0], [2.0, 3.0]]);
+    let hs = t.index_select(&ids.to_dtype(DType::I64)?, 0)?;
+    assert_eq!(hs.to_vec2::<f32>()?, &[[0.0, 1.0], [4.0, 5.0], [2.0, 3.0]]);
     Ok(())
 }
 
@@ -734,44 +747,47 @@ fn index_select(device: &Device) -> Result<()> {
             [9.0, 10.0, 11.0]
         ]
     );
-    let hs = t.index_select(&ids, 1)?;
-    assert_eq!(
-        hs.to_vec2::<f32>()?,
-        &[
-            [0.0, 2.0, 1.0],
-            [3.0, 5.0, 4.0],
-            [6.0, 8.0, 7.0],
-            [9.0, 11.0, 10.0]
-        ]
-    );
-    let hs = t.index_select(&ids, 0)?;
-    assert_eq!(
-        hs.to_vec2::<f32>()?,
-        &[[0.0, 1.0, 2.0], [6.0, 7.0, 8.0], [3.0, 4.0, 5.0]]
-    );
-    // Prior to https://github.com/huggingface/candle/pull/1022
-    // There would be a bug where the last values in the result tensor would be set to 0.
-    let ids = Tensor::new(&[0u32, 2u32, 1u32, 0u32, 2u32, 1u32], device)?;
-    let hs = t.index_select(&ids, 0)?;
-    assert_eq!(
-        hs.to_vec2::<f32>()?,
-        &[
-            [0.0, 1.0, 2.0],
-            [6.0, 7.0, 8.0],
-            [3.0, 4.0, 5.0],
-            [0.0, 1.0, 2.0],
-            [6.0, 7.0, 8.0],
-            [3.0, 4.0, 5.0],
-        ]
-    );
+    for dtype in [DType::U8, DType::U32, DType::I64] {
+        let ids = ids.to_dtype(dtype)?;
+        let hs = t.index_select(&ids, 1)?;
+        assert_eq!(
+            hs.to_vec2::<f32>()?,
+            &[
+                [0.0, 2.0, 1.0],
+                [3.0, 5.0, 4.0],
+                [6.0, 8.0, 7.0],
+                [9.0, 11.0, 10.0]
+            ]
+        );
+        let hs = t.index_select(&ids, 0)?;
+        assert_eq!(
+            hs.to_vec2::<f32>()?,
+            &[[0.0, 1.0, 2.0], [6.0, 7.0, 8.0], [3.0, 4.0, 5.0]]
+        );
+        // Prior to https://github.com/huggingface/candle/pull/1022
+        // There would be a bug where the last values in the result tensor would be set to 0.
+        let ids = Tensor::new(&[0u32, 2u32, 1u32, 0u32, 2u32, 1u32], device)?;
+        let hs = t.index_select(&ids, 0)?;
+        assert_eq!(
+            hs.to_vec2::<f32>()?,
+            &[
+                [0.0, 1.0, 2.0],
+                [6.0, 7.0, 8.0],
+                [3.0, 4.0, 5.0],
+                [0.0, 1.0, 2.0],
+                [6.0, 7.0, 8.0],
+                [3.0, 4.0, 5.0],
+            ]
+        );
 
-    // Test when selecting dim > 0 with ids size different from elem count of
-    // target dim in source/input.
-    let ids = Tensor::new(&[1u32, 0u32, 1u32], device)?;
-    let t = Tensor::arange(1f32, 5f32, device)?.reshape((2, 2))?;
-    assert_eq!(t.to_vec2::<f32>()?, &[[1.0, 2.0], [3.0, 4.0]]);
-    let hs = t.index_select(&ids, 1)?;
-    assert_eq!(hs.to_vec2::<f32>()?, &[[2.0, 1.0, 2.0], [4.0, 3.0, 4.0]]);
+        // Test when selecting dim > 0 with ids size different from elem count of
+        // target dim in source/input.
+        let ids = Tensor::new(&[1u32, 0u32, 1u32], device)?;
+        let t = Tensor::arange(1f32, 5f32, device)?.reshape((2, 2))?;
+        assert_eq!(t.to_vec2::<f32>()?, &[[1.0, 2.0], [3.0, 4.0]]);
+        let hs = t.index_select(&ids, 1)?;
+        assert_eq!(hs.to_vec2::<f32>()?, &[[2.0, 1.0, 2.0], [4.0, 3.0, 4.0]]);
+    }
 
     Ok(())
 }
@@ -933,74 +949,6 @@ fn gather(device: &Device) -> Result<()> {
     Ok(())
 }
 
-fn matmul(device: &Device) -> Result<()> {
-    let data = vec![1.0f32, 2.0, 3.0, 4.0];
-    let a = Tensor::from_slice(&data, (2, 2), device)?;
-    let data = vec![1.0f32, 2.0, 3.0, 4.0];
-    let b = Tensor::from_slice(&data, (2, 2), device)?;
-
-    let c = a.matmul(&b)?;
-    assert_eq!(c.to_vec2::<f32>()?, &[[7.0f32, 10.0], [15.0, 22.0]]);
-
-    let data = vec![1.0f32, 2.0];
-    let a = Tensor::from_slice(&data, (2, 1), device)?;
-    let data = vec![3.0f32, 4.0];
-    let b = Tensor::from_slice(&data, (1, 2), device)?;
-    let c = a.matmul(&b)?;
-    assert_eq!(c.to_vec2::<f32>()?, &[&[3.0, 4.0], &[6.0, 8.0]]);
-
-    let data: Vec<_> = (0..6).map(|i| i as f32).collect();
-    let a = Tensor::from_slice(&data, (2, 3), device)?;
-    let data: Vec<_> = (0..6).map(|i| (i + 2) as f32).collect();
-    let b = Tensor::from_slice(&data, (3, 2), device)?;
-    let c = a.matmul(&b)?;
-    assert_eq!(c.to_vec2::<f32>()?, &[&[16., 19.], &[52., 64.]]);
-
-    let data: Vec<_> = (0..12).map(|i| i as f32).collect();
-    let a = Tensor::from_slice(&data, (2, 2, 3), device)?;
-    let data: Vec<_> = (0..12).map(|i| (i + 2) as f32).collect();
-    let b = Tensor::from_slice(&data, (2, 3, 2), device)?;
-    let expected = [[[16., 19.], [52., 64.]], [[214., 235.], [304., 334.]]];
-
-    let c = a.matmul(&b)?;
-    assert_eq!(c.to_vec3::<f32>()?, &expected);
-
-    // Also perform the matmul on contiguous transposed versions.
-    let a_tt = a.t()?.contiguous()?.t()?;
-    assert!(!a_tt.is_contiguous());
-    assert_eq!(a.dims(), a_tt.dims());
-    assert_eq!(a_tt.stride(), &[6, 1, 2]);
-
-    let b_tt = b.t()?.contiguous()?.t()?;
-    assert!(!b_tt.is_contiguous());
-    assert_eq!(b.dims(), b_tt.dims());
-    assert_eq!(b_tt.stride(), &[6, 1, 3]);
-
-    assert_eq!(a_tt.matmul(&b)?.to_vec3::<f32>()?, &expected);
-    assert_eq!(a.matmul(&b_tt)?.to_vec3::<f32>()?, &expected);
-    assert_eq!(a_tt.matmul(&b_tt)?.to_vec3::<f32>()?, &expected);
-    Ok(())
-}
-
-fn broadcast_matmul(device: &Device) -> Result<()> {
-    let lhs = Tensor::randn(0f32, 1f32, (3, 1, 4, 5), device)?;
-    let rhs = Tensor::randn(0f32, 1f32, (6, 5, 2), device)?;
-    let out = lhs.broadcast_matmul(&rhs)?;
-    assert_eq!(out.dims(), &[3, 6, 4, 2]);
-    for idx1 in 0..3 {
-        for idx2 in 0..6 {
-            let out = out.i((idx1, idx2))?;
-            let lhs = lhs.i((idx1, 0))?;
-            let rhs = rhs.i(idx2)?;
-            let out2 = lhs.matmul(&rhs);
-            let sum_diff2 = (out - out2)?.sqr()?.sum_all()?;
-            // With cuda, we see errors of up to ~1e-12.
-            assert!(sum_diff2.to_vec0::<f32>()? < 1e-6)
-        }
-    }
-    Ok(())
-}
-
 fn broadcasting(device: &Device) -> Result<()> {
     let t1 = Tensor::arange(0f32, 24f32, device)?.reshape((4, 2, 3))?;
     let t2 = Tensor::new(&[100f32, 200f32], device)?;
@@ -1135,6 +1083,27 @@ fn randn(device: &Device) -> Result<()> {
     Ok(())
 }
 
+fn zero_dim(device: &Device) -> Result<()> {
+    let t = Tensor::zeros((4, 0, 1), DType::F32, device)?;
+    assert_eq!(t.dims3()?, (4, 0, 1));
+    let t2 = Tensor::zeros((4, 3, 1), DType::F32, device)?;
+    let t_cat = Tensor::cat(&[&t, &t2], 1)?;
+    assert_eq!(t_cat.dims3()?, (4, 3, 1));
+    let t_cat = Tensor::cat(&[&t, &t], 1)?;
+    assert_eq!(t_cat.dims3()?, (4, 0, 1));
+    let t_unary = t.sqrt()?;
+    assert_eq!(t_unary.dims3()?, (4, 0, 1));
+    let t_plus = (&t + 1.)?;
+    assert_eq!(t_plus.dims3()?, (4, 0, 1));
+    let t_mm = t2.matmul(&t.t()?)?;
+    assert_eq!(t_mm.dims3()?, (4, 3, 0));
+    let t_mm = t.matmul(&t2.t()?)?;
+    assert_eq!(t_mm.dims3()?, (4, 0, 3));
+    let t_mm = t.t()?.matmul(&t)?;
+    assert_eq!(t_mm.dims3()?, (4, 1, 1));
+    Ok(())
+}
+
 test_device!(zeros, zeros_cpu, zeros_gpu, zeros_metal);
 test_device!(ones, ones_cpu, ones_gpu, ones_metal);
 test_device!(full, full_cpu, full_gpu, full_metal);
@@ -1154,13 +1123,6 @@ test_device!(unary_op, unary_op_cpu, unary_op_gpu, unary_op_metal);
 test_device!(binary_op, binary_op_cpu, binary_op_gpu, binary_op_metal);
 test_device!(embeddings, embeddings_cpu, embeddings_gpu, embeddings_metal);
 test_device!(cmp, cmp_cpu, cmp_gpu, cmp_metal);
-test_device!(matmul, matmul_cpu, matmul_gpu, matmul_metal);
-test_device!(
-    broadcast_matmul,
-    broadcast_matmul_cpu,
-    broadcast_matmul_gpu,
-    broadcast_matmul_metal
-);
 test_device!(
     broadcasting,
     broadcasting_cpu,
@@ -1190,6 +1152,7 @@ test_device!(
 test_device!(randn, randn_cpu, randn_gpu, randn_metal);
 test_device!(clamp, clamp_cpu, clamp_gpu, clamp_metal);
 test_device!(var, var_cpu, var_gpu, var_metal);
+test_device!(zero_dim, zero_dim_cpu, zero_dim_gpu, zero_dim_metal);
 
 // There was originally a bug on the CPU implementation for randn
 // https://github.com/huggingface/candle/issues/381
@@ -1317,8 +1280,8 @@ fn pow() -> Result<()> {
     let rhs = (&lhs - 2.)?;
     let res = lhs.pow(&rhs)?;
     assert_eq!(
-        test_utils::to_vec2_round(&res, 4)?,
-        [[1.0, 1.0, 3.0], [16.0, 125.0, 1296.0001]]
+        test_utils::to_vec2_round(&res, 3)?,
+        [[1.0, 1.0, 3.0], [16.0, 125.0, 1296.0]]
     );
     Ok(())
 }

candle-examples/Cargo.toml

@@ -25,8 +25,9 @@ hf-hub = { workspace = true, features = ["tokio"] }
 image = { workspace = true }
 intel-mkl-src = { workspace = true, optional = true }
 num-traits = { workspace = true }
-pyo3 = { version = "0.20.0", features = ["auto-initialize"], optional = true }
+pyo3 = { version = "0.21.0", features = ["auto-initialize"], optional = true }
 rayon = { workspace = true }
+rubato = { version = "0.15.0", optional = true }
 safetensors = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
@@ -41,7 +42,7 @@ clap = { workspace = true }
 imageproc = { workspace = true }
 memmap2 = { workspace = true }
 rand = { workspace = true }
-rusttype = { workspace = true }
+ab_glyph = { workspace = true }
 tracing = { workspace = true }
 tracing-chrome = { workspace = true }
 tracing-subscriber = { workspace = true }
@@ -63,6 +64,7 @@ nccl = ["cuda", "cudarc/nccl", "dep:half"]
 onnx = ["candle-onnx"]
 metal = ["candle/metal", "candle-nn/metal"]
 microphone = ["cpal"]
+encodec = ["cpal", "symphonia", "rubato"]
 
 [[example]]
 name = "llama_multiprocess"
@@ -98,6 +100,4 @@ required-features = ["candle-datasets"]
 
 [[example]]
 name = "encodec"
-required-features = ["symphonia"]
-
-
+required-features = ["encodec"]

candle-examples/examples/clip/README.md

@@ -0,0 +1,46 @@
+Contrastive Language-Image Pre-Training
+
+Contrastive Language-Image Pre-Training (CLIP) is an architecture trained on
+pairs of images with related texts.
+
+https://github.com/openai/CLIP
+
+https://github.com/huggingface/transformers/tree/f6fa0f0bf0796ac66f201f23bdb8585de1609add/src/transformers/models/clip
+
+## Running on an example on cpu
+
+```
+$ cargo run --example clip --release -- --images "candle-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg","candle-examples/examples/yolo-v8/assets/bike.jpg" --cpu --sequences  "a cycling race","a photo of two cats","a robot holding a candle"
+
+
+Results for image: candle-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg
+
+INFO clip: Probability: 0.0000% Text: a cycling race
+INFO clip: Probability: 0.0000% Text: a photo of two cats
+INFO clip: Probability: 100.0000% Text: a robot holding a candle
+
+Results for image: candle-examples/examples/yolo-v8/assets/bike.jpg
+
+INFO clip: Probability: 99.9999% Text: a cycling race
+INFO clip: Probability: 0.0001% Text: a photo of two cats
+INFO clip: Probability: 0.0000% Text: a robot holding a candle
+```
+
+## Running on an example with metal feature (mac)
+
+```
+$ cargo run --features metal --example clip --release -- --images "candle-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg","candle-examples/examples/yolo-v8/assets/bike.jpg" --cpu --sequences "a cycling race","a photo of two cats","a robot holding a candle"
+
+
+Results for image: candle-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg
+
+INFO clip: Probability: 0.0000% Text: a cycling race
+INFO clip: Probability: 0.0000% Text: a photo of two cats
+INFO clip: Probability: 100.0000% Text: a robot holding a candle
+
+Results for image: candle-examples/examples/yolo-v8/assets/bike.jpg
+
+INFO clip: Probability: 99.9999% Text: a cycling race
+INFO clip: Probability: 0.0001% Text: a photo of two cats
+INFO clip: Probability: 0.0000% Text: a robot holding a candle
+```

candle-examples/examples/clip/main.rs

@@ -0,0 +1,202 @@
+#[cfg(feature = "mkl")]
+extern crate intel_mkl_src;
+
+#[cfg(feature = "accelerate")]
+extern crate accelerate_src;
+
+use anyhow::Error as E;
+use clap::Parser;
+
+use candle::{DType, Device, Tensor};
+use candle_nn::{ops::softmax, VarBuilder};
+use candle_transformers::models::clip;
+
+use tokenizers::Tokenizer;
+use tracing::info;
+
+#[derive(Parser)]
+struct Args {
+    #[arg(long)]
+    model: Option<String>,
+
+    #[arg(long)]
+    tokenizer: Option<String>,
+
+    #[arg(long, use_value_delimiter = true)]
+    images: Option<Vec<String>>,
+
+    #[arg(long)]
+    cpu: bool,
+
+    #[arg(long, use_value_delimiter = true)]
+    sequences: Option<Vec<String>>,
+}
+
+fn load_image<T: AsRef<std::path::Path>>(path: T, image_size: usize) -> anyhow::Result<Tensor> {
+    let img = image::io::Reader::open(path)?.decode()?;
+    let (height, width) = (image_size, image_size);
+    let img = img.resize_to_fill(
+        width as u32,
+        height as u32,
+        image::imageops::FilterType::Triangle,
+    );
+
+    let img = img.to_rgb8();
+
+    let img = img.into_raw();
+    let img = Tensor::from_vec(img, (height, width, 3), &Device::Cpu)?
+        .permute((2, 0, 1))?
+        .to_dtype(DType::F32)?
+        .affine(2. / 255., -1.)?;
+    // .unsqueeze(0)?;
+    Ok(img)
+}
+
+fn load_images<T: AsRef<std::path::Path>>(
+    paths: &Vec<T>,
+    image_size: usize,
+) -> anyhow::Result<Tensor> {
+    let mut images = vec![];
+
+    for path in paths {
+        let tensor = load_image(path, image_size)?;
+        images.push(tensor);
+    }
+
+    let images = Tensor::stack(&images, 0)?;
+
+    Ok(images)
+}
+
+pub fn main() -> anyhow::Result<()> {
+    // std::env::set_var("RUST_BACKTRACE", "full");
+
+    let args = Args::parse();
+
+    tracing_subscriber::fmt::init();
+
+    let model_file = match args.model {
+        None => {
+            let api = hf_hub::api::sync::Api::new()?;
+
+            let api = api.repo(hf_hub::Repo::with_revision(
+                "openai/clip-vit-base-patch32".to_string(),
+                hf_hub::RepoType::Model,
+                "refs/pr/15".to_string(),
+            ));
+
+            api.get("model.safetensors")?
+        }
+        Some(model) => model.into(),
+    };
+
+    let tokenizer = get_tokenizer(args.tokenizer)?;
+
+    let config = clip::ClipConfig::vit_base_patch32();
+
+    let device = candle_examples::device(args.cpu)?;
+
+    let vec_imgs = match args.images {
+        Some(imgs) => imgs,
+        None => vec![
+            "candle-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg".to_string(),
+            "candle-examples/examples/yolo-v8/assets/bike.jpg".to_string(),
+        ],
+    };
+
+    // let image = load_image(args.image, config.image_size)?.to_device(&device)?;
+    let images = load_images(&vec_imgs, config.image_size)?.to_device(&device)?;
+
+    let vb =
+        unsafe { VarBuilder::from_mmaped_safetensors(&[model_file.clone()], DType::F32, &device)? };
+
+    let model = clip::ClipModel::new(vb, &config)?;
+
+    let (input_ids, vec_seq) = tokenize_sequences(args.sequences, &tokenizer, &device)?;
+
+    let (_logits_per_text, logits_per_image) = model.forward(&images, &input_ids)?;
+
+    let softmax_image = softmax(&logits_per_image, 1)?;
+
+    let softmax_image_vec = softmax_image.flatten_all()?.to_vec1::<f32>()?;
+
+    info!("softmax_image_vec: {:?}", softmax_image_vec);
+
+    let probability_vec = softmax_image_vec
+        .iter()
+        .map(|v| v * 100.0)
+        .collect::<Vec<f32>>();
+
+    let probability_per_image = probability_vec.len() / vec_imgs.len();
+
+    for (i, img) in vec_imgs.iter().enumerate() {
+        let start = i * probability_per_image;
+        let end = start + probability_per_image;
+        let prob = &probability_vec[start..end];
+        info!("\n\nResults for image: {}\n", img);
+
+        for (i, p) in prob.iter().enumerate() {
+            info!("Probability: {:.4}% Text: {} ", p, vec_seq[i]);
+        }
+    }
+
+    Ok(())
+}
+
+pub fn get_tokenizer(tokenizer: Option<String>) -> anyhow::Result<Tokenizer> {
+    let tokenizer = match tokenizer {
+        None => {
+            let api = hf_hub::api::sync::Api::new()?;
+            let api = api.repo(hf_hub::Repo::with_revision(
+                "openai/clip-vit-base-patch32".to_string(),
+                hf_hub::RepoType::Model,
+                "refs/pr/15".to_string(),
+            ));
+            api.get("tokenizer.json")?
+        }
+        Some(file) => file.into(),
+    };
+
+    Tokenizer::from_file(tokenizer).map_err(E::msg)
+}
+
+pub fn tokenize_sequences(
+    sequences: Option<Vec<String>>,
+    tokenizer: &Tokenizer,
+    device: &Device,
+) -> anyhow::Result<(Tensor, Vec<String>)> {
+    let pad_id = *tokenizer
+        .get_vocab(true)
+        .get("<|endoftext|>")
+        .ok_or(E::msg("No pad token"))?;
+
+    let vec_seq = match sequences {
+        Some(seq) => seq,
+        None => vec![
+            "a cycling race".to_string(),
+            "a photo of two cats".to_string(),
+            "a robot holding a candle".to_string(),
+        ],
+    };
+
+    let mut tokens = vec![];
+
+    for seq in vec_seq.clone() {
+        let encoding = tokenizer.encode(seq, true).map_err(E::msg)?;
+        tokens.push(encoding.get_ids().to_vec());
+    }
+
+    let max_len = tokens.iter().map(|v| v.len()).max().unwrap_or(0);
+
+    // Pad the sequences to have the same length
+    for token_vec in tokens.iter_mut() {
+        let len_diff = max_len - token_vec.len();
+        if len_diff > 0 {
+            token_vec.extend(vec![pad_id; len_diff]);
+        }
+    }
+
+    let input_ids = Tensor::new(tokens, device)?;
+
+    Ok((input_ids, vec_seq))
+}

candle-examples/examples/encodec/README.md

@@ -13,8 +13,13 @@ cargo run --example encodec --features symphonia --release -- code-to-audio \
 ```
 
 This decodes the EnCodec tokens stored in `jfk-codes.safetensors` and generates
-an output wav file containing the audio data. Instead of `code-to-audio` one
-can use:
+an output wav file containing the audio data.
+
+Instead of `code-to-audio` one can use:
 - `audio-to-audio in.mp3 out.wav`: encodes the input audio file then decodes it to a wav file.
 - `audio-to-code in.mp3 out.safetensors`: generates a safetensors file
   containing EnCodec tokens for the input audio file.
+
+If the audio output file name is set to `-`, the audio content directly gets
+played on default audio output device. If the audio input file is set to `-`, the audio
+gets recorded from the default audio input.

candle-examples/examples/encodec/audio_io.rs

@@ -0,0 +1,275 @@
+#![allow(unused)]
+use anyhow::{Context, Result};
+use std::sync::{Arc, Mutex};
+
+pub const SAMPLE_RATE: usize = 24_000;
+
+pub(crate) struct AudioOutputData_ {
+    resampled_data: std::collections::VecDeque<f32>,
+    resampler: rubato::FastFixedIn<f32>,
+    output_buffer: Vec<f32>,
+    input_buffer: Vec<f32>,
+    input_len: usize,
+}
+
+impl AudioOutputData_ {
+    pub(crate) fn new(input_sample_rate: usize, output_sample_rate: usize) -> Result<Self> {
+        use rubato::Resampler;
+
+        let resampled_data = std::collections::VecDeque::with_capacity(output_sample_rate * 10);
+        let resample_ratio = output_sample_rate as f64 / input_sample_rate as f64;
+        let resampler = rubato::FastFixedIn::new(
+            resample_ratio,
+            f64::max(resample_ratio, 1.0),
+            rubato::PolynomialDegree::Septic,
+            1024,
+            1,
+        )?;
+        let input_buffer = resampler.input_buffer_allocate(true).remove(0);
+        let output_buffer = resampler.output_buffer_allocate(true).remove(0);
+        Ok(Self {
+            resampled_data,
+            resampler,
+            input_buffer,
+            output_buffer,
+            input_len: 0,
+        })
+    }
+
+    pub fn reset(&mut self) {
+        use rubato::Resampler;
+        self.output_buffer.fill(0.);
+        self.input_buffer.fill(0.);
+        self.resampler.reset();
+        self.resampled_data.clear();
+    }
+
+    pub(crate) fn take_all(&mut self) -> Vec<f32> {
+        let mut data = Vec::with_capacity(self.resampled_data.len());
+        while let Some(elem) = self.resampled_data.pop_back() {
+            data.push(elem);
+        }
+        data
+    }
+
+    pub(crate) fn is_empty(&self) -> bool {
+        self.resampled_data.is_empty()
+    }
+
+    // Assumes that the input buffer is large enough.
+    fn push_input_buffer(&mut self, samples: &[f32]) {
+        self.input_buffer[self.input_len..self.input_len + samples.len()].copy_from_slice(samples);
+        self.input_len += samples.len()
+    }
+
+    pub(crate) fn push_samples(&mut self, samples: &[f32]) -> Result<()> {
+        use rubato::Resampler;
+
+        let mut pos_in = 0;
+        loop {
+            let rem = self.input_buffer.len() - self.input_len;
+            let pos_end = usize::min(pos_in + rem, samples.len());
+            self.push_input_buffer(&samples[pos_in..pos_end]);
+            pos_in = pos_end;
+            if self.input_len < self.input_buffer.len() {
+                break;
+            }
+            let (_, out_len) = self.resampler.process_into_buffer(
+                &[&self.input_buffer],
+                &mut [&mut self.output_buffer],
+                None,
+            )?;
+            for &elem in self.output_buffer[..out_len].iter() {
+                self.resampled_data.push_front(elem)
+            }
+            self.input_len = 0;
+        }
+        Ok(())
+    }
+}
+
+type AudioOutputData = Arc<Mutex<AudioOutputData_>>;
+
+pub(crate) fn setup_output_stream() -> Result<(cpal::Stream, AudioOutputData)> {
+    use cpal::traits::{DeviceTrait, HostTrait, StreamTrait};
+
+    println!("Setup audio output stream!");
+    let host = cpal::default_host();
+    let device = host
+        .default_output_device()
+        .context("no output device available")?;
+    let mut supported_configs_range = device.supported_output_configs()?;
+    let config_range = match supported_configs_range.find(|c| c.channels() == 1) {
+        // On macOS, it's commonly the case that there are only stereo outputs.
+        None => device
+            .supported_output_configs()?
+            .next()
+            .context("no audio output available")?,
+        Some(config_range) => config_range,
+    };
+    let sample_rate = cpal::SampleRate(SAMPLE_RATE as u32).clamp(
+        config_range.min_sample_rate(),
+        config_range.max_sample_rate(),
+    );
+    let config: cpal::StreamConfig = config_range.with_sample_rate(sample_rate).into();
+    let channels = config.channels as usize;
+    println!(
+        "cpal device: {} {} {config:?}",
+        device.name().unwrap_or_else(|_| "unk".to_string()),
+        config.sample_rate.0
+    );
+    let audio_data = Arc::new(Mutex::new(AudioOutputData_::new(
+        SAMPLE_RATE,
+        config.sample_rate.0 as usize,
+    )?));
+    let ad = audio_data.clone();
+    let stream = device.build_output_stream(
+        &config,
+        move |data: &mut [f32], _: &cpal::OutputCallbackInfo| {
+            data.fill(0.);
+            let mut ad = ad.lock().unwrap();
+            let mut last_elem = 0f32;
+            for (idx, elem) in data.iter_mut().enumerate() {
+                if idx % channels == 0 {
+                    match ad.resampled_data.pop_back() {
+                        None => break,
+                        Some(v) => {
+                            last_elem = v;
+                            *elem = v
+                        }
+                    }
+                } else {
+                    *elem = last_elem
+                }
+            }
+        },
+        move |err| eprintln!("cpal error: {err}"),
+        None, // None=blocking, Some(Duration)=timeout
+    )?;
+    stream.play()?;
+    Ok((stream, audio_data))
+}
+
+pub(crate) fn setup_input_stream() -> Result<(cpal::Stream, AudioOutputData)> {
+    use cpal::traits::{DeviceTrait, HostTrait, StreamTrait};
+
+    println!("Setup audio input stream!");
+    let host = cpal::default_host();
+    let device = host
+        .default_input_device()
+        .context("no input device available")?;
+    let mut supported_configs_range = device.supported_input_configs()?;
+    let config_range = supported_configs_range
+        .find(|c| c.channels() == 1)
+        .context("no audio input available")?;
+    let sample_rate = cpal::SampleRate(SAMPLE_RATE as u32).clamp(
+        config_range.min_sample_rate(),
+        config_range.max_sample_rate(),
+    );
+    let config: cpal::StreamConfig = config_range.with_sample_rate(sample_rate).into();
+    println!(
+        "cpal device: {} {} {config:?}",
+        device.name().unwrap_or_else(|_| "unk".to_string()),
+        config.sample_rate.0
+    );
+    let audio_data = Arc::new(Mutex::new(AudioOutputData_::new(
+        config.sample_rate.0 as usize,
+        SAMPLE_RATE,
+    )?));
+    let ad = audio_data.clone();
+    let stream = device.build_input_stream(
+        &config,
+        move |data: &[f32], _: &cpal::InputCallbackInfo| {
+            let mut ad = ad.lock().unwrap();
+            if let Err(err) = ad.push_samples(data) {
+                eprintln!("error processing audio input {err:?}")
+            }
+        },
+        move |err| eprintln!("cpal error: {err}"),
+        None, // None=blocking, Some(Duration)=timeout
+    )?;
+    stream.play()?;
+    Ok((stream, audio_data))
+}
+
+fn conv<T>(samples: &mut Vec<f32>, data: std::borrow::Cow<symphonia::core::audio::AudioBuffer<T>>)
+where
+    T: symphonia::core::sample::Sample,
+    f32: symphonia::core::conv::FromSample<T>,
+{
+    use symphonia::core::audio::Signal;
+    use symphonia::core::conv::FromSample;
+    samples.extend(data.chan(0).iter().map(|v| f32::from_sample(*v)))
+}
+
+pub(crate) fn pcm_decode<P: AsRef<std::path::Path>>(path: P) -> Result<(Vec<f32>, u32)> {
+    use symphonia::core::audio::{AudioBufferRef, Signal};
+
+    let src = std::fs::File::open(path)?;
+    let mss = symphonia::core::io::MediaSourceStream::new(Box::new(src), Default::default());
+    let hint = symphonia::core::probe::Hint::new();
+    let meta_opts: symphonia::core::meta::MetadataOptions = Default::default();
+    let fmt_opts: symphonia::core::formats::FormatOptions = Default::default();
+    let probed = symphonia::default::get_probe().format(&hint, mss, &fmt_opts, &meta_opts)?;
+    let mut format = probed.format;
+    let track = format
+        .tracks()
+        .iter()
+        .find(|t| t.codec_params.codec != symphonia::core::codecs::CODEC_TYPE_NULL)
+        .expect("no supported audio tracks");
+    let mut decoder = symphonia::default::get_codecs()
+        .make(&track.codec_params, &Default::default())
+        .expect("unsupported codec");
+    let track_id = track.id;
+    let sample_rate = track.codec_params.sample_rate.unwrap_or(0);
+    let mut pcm_data = Vec::new();
+    while let Ok(packet) = format.next_packet() {
+        while !format.metadata().is_latest() {
+            format.metadata().pop();
+        }
+        if packet.track_id() != track_id {
+            continue;
+        }
+        match decoder.decode(&packet)? {
+            AudioBufferRef::F32(buf) => pcm_data.extend(buf.chan(0)),
+            AudioBufferRef::U8(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::U16(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::U24(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::U32(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::S8(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::S16(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::S24(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::S32(data) => conv(&mut pcm_data, data),
+            AudioBufferRef::F64(data) => conv(&mut pcm_data, data),
+        }
+    }
+    Ok((pcm_data, sample_rate))
+}
+
+pub(crate) fn resample(pcm_in: &[f32], sr_in: usize, sr_out: usize) -> Result<Vec<f32>> {
+    use rubato::Resampler;
+
+    let mut pcm_out =
+        Vec::with_capacity((pcm_in.len() as f64 * sr_out as f64 / sr_in as f64) as usize + 1024);
+
+    let mut resampler = rubato::FftFixedInOut::<f32>::new(sr_in, sr_out, 1024, 1)?;
+    let mut output_buffer = resampler.output_buffer_allocate(true);
+    let mut pos_in = 0;
+    while pos_in + resampler.input_frames_next() < pcm_in.len() {
+        let (in_len, out_len) =
+            resampler.process_into_buffer(&[&pcm_in[pos_in..]], &mut output_buffer, None)?;
+        pos_in += in_len;
+        pcm_out.extend_from_slice(&output_buffer[0][..out_len]);
+    }
+
+    if pos_in < pcm_in.len() {
+        let (_in_len, out_len) = resampler.process_partial_into_buffer(
+            Some(&[&pcm_in[pos_in..]]),
+            &mut output_buffer,
+            None,
+        )?;
+        pcm_out.extend_from_slice(&output_buffer[0][..out_len]);
+    }
+
+    Ok(pcm_out)
+}

candle-examples/examples/encodec/main.rs

@@ -11,59 +11,7 @@ use candle_transformers::models::encodec::{Config, Model};
 use clap::{Parser, ValueEnum};
 use hf_hub::api::sync::Api;
 
-fn conv<T>(samples: &mut Vec<f32>, data: std::borrow::Cow<symphonia::core::audio::AudioBuffer<T>>)
-where
-    T: symphonia::core::sample::Sample,
-    f32: symphonia::core::conv::FromSample<T>,
-{
-    use symphonia::core::audio::Signal;
-    use symphonia::core::conv::FromSample;
-    samples.extend(data.chan(0).iter().map(|v| f32::from_sample(*v)))
-}
-
-fn pcm_decode<P: AsRef<std::path::Path>>(path: P) -> anyhow::Result<(Vec<f32>, u32)> {
-    use symphonia::core::audio::{AudioBufferRef, Signal};
-
-    let src = std::fs::File::open(path)?;
-    let mss = symphonia::core::io::MediaSourceStream::new(Box::new(src), Default::default());
-    let hint = symphonia::core::probe::Hint::new();
-    let meta_opts: symphonia::core::meta::MetadataOptions = Default::default();
-    let fmt_opts: symphonia::core::formats::FormatOptions = Default::default();
-    let probed = symphonia::default::get_probe().format(&hint, mss, &fmt_opts, &meta_opts)?;
-    let mut format = probed.format;
-    let track = format
-        .tracks()
-        .iter()
-        .find(|t| t.codec_params.codec != symphonia::core::codecs::CODEC_TYPE_NULL)
-        .expect("no supported audio tracks");
-    let mut decoder = symphonia::default::get_codecs()
-        .make(&track.codec_params, &Default::default())
-        .expect("unsupported codec");
-    let track_id = track.id;
-    let sample_rate = track.codec_params.sample_rate.unwrap_or(0);
-    let mut pcm_data = Vec::new();
-    while let Ok(packet) = format.next_packet() {
-        while !format.metadata().is_latest() {
-            format.metadata().pop();
-        }
-        if packet.track_id() != track_id {
-            continue;
-        }
-        match decoder.decode(&packet)? {
-            AudioBufferRef::F32(buf) => pcm_data.extend(buf.chan(0)),
-            AudioBufferRef::U8(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::U16(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::U24(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::U32(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::S8(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::S16(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::S24(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::S32(data) => conv(&mut pcm_data, data),
-            AudioBufferRef::F64(data) => conv(&mut pcm_data, data),
-        }
-    }
-    Ok((pcm_data, sample_rate))
-}
+mod audio_io;
 
 #[derive(Clone, Debug, Copy, PartialEq, Eq, ValueEnum)]
 enum Action {
@@ -109,14 +57,36 @@ fn main() -> Result<()> {
     let codes = match args.action {
         Action::CodeToAudio => {
             let codes = candle::safetensors::load(args.in_file, &device)?;
-            let codes = codes.get("codes").expect("no codes in input file").i(0)?;
-            codes
+            codes.get("codes").expect("no codes in input file").clone()
         }
         Action::AudioToCode | Action::AudioToAudio => {
-            let (pcm, sample_rate) = pcm_decode(args.in_file)?;
-            if sample_rate != 24_000 {
-                println!("WARNING: encodec uses a 24khz sample rate, input uses {sample_rate}")
-            }
+            let pcm = if args.in_file == "-" {
+                println!(">>>> RECORDING AUDIO, PRESS ENTER ONCE DONE <<<<");
+                let (stream, input_audio) = audio_io::setup_input_stream()?;
+                let mut pcms = vec![];
+                let stdin = std::thread::spawn(|| {
+                    let mut s = String::new();
+                    std::io::stdin().read_line(&mut s)
+                });
+                while !stdin.is_finished() {
+                    let input = input_audio.lock().unwrap().take_all();
+                    if input.is_empty() {
+                        std::thread::sleep(std::time::Duration::from_millis(100));
+                        continue;
+                    }
+                    pcms.push(input)
+                }
+                drop(stream);
+                pcms.concat()
+            } else {
+                let (pcm, sample_rate) = audio_io::pcm_decode(args.in_file)?;
+                if sample_rate != 24_000 {
+                    println!("WARNING: encodec uses a 24khz sample rate, input uses {sample_rate}, resampling...");
+                    audio_io::resample(&pcm, sample_rate as usize, 24_000)?
+                } else {
+                    pcm
+                }
+            };
             let pcm_len = pcm.len();
             let pcm = Tensor::from_vec(pcm, (1, 1, pcm_len), &device)?;
             println!("input pcm shape: {:?}", pcm.shape());
@@ -135,8 +105,26 @@ fn main() -> Result<()> {
             let pcm = pcm.i(0)?.i(0)?;
             let pcm = candle_examples::audio::normalize_loudness(&pcm, 24_000, true)?;
             let pcm = pcm.to_vec1::<f32>()?;
-            let mut output = std::fs::File::create(&args.out_file)?;
-            candle_examples::wav::write_pcm_as_wav(&mut output, &pcm, 24_000)?;
+            if args.out_file == "-" {
+                let (stream, ad) = audio_io::setup_output_stream()?;
+                {
+                    let mut ad = ad.lock().unwrap();
+                    ad.push_samples(&pcm)?;
+                }
+                loop {
+                    let ad = ad.lock().unwrap();
+                    if ad.is_empty() {
+                        break;
+                    }
+                    // That's very weird, calling thread::sleep here triggers the stream to stop
+                    // playing (the callback doesn't seem to be called anymore).
+                    // std::thread::sleep(std::time::Duration::from_millis(100));
+                }
+                drop(stream)
+            } else {
+                let mut output = std::fs::File::create(&args.out_file)?;
+                candle_examples::wav::write_pcm_as_wav(&mut output, &pcm, 24_000)?;
+            }
         }
     }
     Ok(())

candle-examples/examples/gemma/main.rs

@@ -16,6 +16,30 @@ use candle_transformers::generation::LogitsProcessor;
 use hf_hub::{api::sync::Api, Repo, RepoType};
 use tokenizers::Tokenizer;
 
+#[derive(Clone, Debug, Copy, PartialEq, Eq, clap::ValueEnum)]
+enum Which {
+    #[value(name = "2b")]
+    Base2B,
+    #[value(name = "7b")]
+    Base7B,
+    #[value(name = "2b-it")]
+    Instruct2B,
+    #[value(name = "7b-it")]
+    Instruct7B,
+    #[value(name = "1.1-2b-it")]
+    InstructV1_1_2B,
+    #[value(name = "1.1-7b-it")]
+    InstructV1_1_7B,
+    #[value(name = "code-2b")]
+    CodeBase2B,
+    #[value(name = "code-7b")]
+    CodeBase7B,
+    #[value(name = "code-2b-it")]
+    CodeInstruct2B,
+    #[value(name = "code-7b-it")]
+    CodeInstruct7B,
+}
+
 struct TextGeneration {
     model: Model,
     device: Device,
@@ -165,6 +189,10 @@ struct Args {
     /// The context size to consider for the repeat penalty.
     #[arg(long, default_value_t = 64)]
     repeat_last_n: usize,
+
+    /// The model to use.
+    #[arg(long, default_value = "2b")]
+    which: Which,
 }
 
 fn main() -> Result<()> {
@@ -196,14 +224,19 @@ fn main() -> Result<()> {
     let start = std::time::Instant::now();
     let api = Api::new()?;
     let model_id = match &args.model_id {
-        Some(model_id) => match model_id.as_str() {
-            "7b-it" => "google/gemma-7b-it".to_string(),
-            "7b" => "google/gemma-7b".to_string(),
-            "2b-it" => "google/gemma-2b-it".to_string(),
-            "2b" => "google/gemma-2b".to_string(),
-            _ => model_id.to_string(),
+        Some(model_id) => model_id.to_string(),
+        None => match args.which {
+            Which::InstructV1_1_2B => "google/gemma-1.1-2b-it".to_string(),
+            Which::InstructV1_1_7B => "google/gemma-1.1-7b-it".to_string(),
+            Which::Base2B => "google/gemma-2b".to_string(),
+            Which::Base7B => "google/gemma-7b".to_string(),
+            Which::Instruct2B => "google/gemma-2b-it".to_string(),
+            Which::Instruct7B => "google/gemma-7b-it".to_string(),
+            Which::CodeBase2B => "google/codegemma-2b".to_string(),
+            Which::CodeBase7B => "google/codegemma-7b".to_string(),
+            Which::CodeInstruct2B => "google/codegemma-2b-it".to_string(),
+            Which::CodeInstruct7B => "google/codegemma-7b-it".to_string(),
         },
-        None => "google/gemma-2b".to_string(),
     };
     let repo = api.repo(Repo::with_revision(
         model_id,

candle-examples/examples/llama/main.rs

@@ -31,6 +31,8 @@ const DEFAULT_PROMPT: &str = "My favorite theorem is ";
 enum Which {
     V1,
     V2,
+    V3,
+    V3Instruct,
     #[value(name = "solar-10.7b")]
     Solar10_7B,
     #[value(name = "tiny-llama-1.1b-chat")]
@@ -45,8 +47,8 @@ struct Args {
     cpu: bool,
 
     /// The temperature used to generate samples.
-    #[arg(long)]
-    temperature: Option<f64>,
+    #[arg(long, default_value_t = 0.8)]
+    temperature: f64,
 
     /// Nucleus sampling probability cutoff.
     #[arg(long)]
@@ -90,11 +92,11 @@ struct Args {
     use_flash_attn: bool,
 
     /// Penalty to be applied for repeating tokens, 1. means no penalty.
-    #[arg(long, default_value_t = 1.0)]
+    #[arg(long, default_value_t = 1.1)]
     repeat_penalty: f32,
 
     /// The context size to consider for the repeat penalty.
-    #[arg(long, default_value_t = 64)]
+    #[arg(long, default_value_t = 128)]
     repeat_last_n: usize,
 }
 
@@ -118,13 +120,18 @@ fn main() -> Result<()> {
         Some("bf16") => DType::BF16,
         Some("f32") => DType::F32,
         Some(dtype) => bail!("Unsupported dtype {dtype}"),
-        None => DType::F16,
+        None => match args.which {
+            Which::V3 | Which::V3Instruct => DType::BF16,
+            Which::V1 | Which::V2 | Which::Solar10_7B | Which::TinyLlama1_1BChat => DType::F16,
+        },
     };
-    let (llama, tokenizer_filename, mut cache) = {
+    let (llama, tokenizer_filename, mut cache, config) = {
         let api = Api::new()?;
         let model_id = args.model_id.unwrap_or_else(|| match args.which {
             Which::V1 => "Narsil/amall-7b".to_string(),
             Which::V2 => "meta-llama/Llama-2-7b-hf".to_string(),
+            Which::V3 => "meta-llama/Meta-Llama-3-8B".to_string(),
+            Which::V3Instruct => "meta-llama/Meta-Llama-3-8B-Instruct".to_string(),
             Which::Solar10_7B => "upstage/SOLAR-10.7B-v1.0".to_string(),
             Which::TinyLlama1_1BChat => "TinyLlama/TinyLlama-1.1B-Chat-v1.0".to_string(),
         });
@@ -138,7 +145,7 @@ fn main() -> Result<()> {
         let config = config.into_config(args.use_flash_attn);
 
         let filenames = match args.which {
-            Which::V1 | Which::V2 | Which::Solar10_7B => {
+            Which::V1 | Which::V2 | Which::V3 | Which::V3Instruct | Which::Solar10_7B => {
                 candle_examples::hub_load_safetensors(&api, "model.safetensors.index.json")?
             }
             Which::TinyLlama1_1BChat => vec![api.get("model.safetensors")?],
@@ -146,10 +153,12 @@ fn main() -> Result<()> {
         let cache = model::Cache::new(!args.no_kv_cache, dtype, &config, &device)?;
 
         let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
-        (Llama::load(vb, &config)?, tokenizer_filename, cache)
+        (Llama::load(vb, &config)?, tokenizer_filename, cache, config)
     };
     let tokenizer = Tokenizer::from_file(tokenizer_filename).map_err(E::msg)?;
-    let eos_token_id = tokenizer.token_to_id(EOS_TOKEN);
+    let eos_token_id = config
+        .eos_token_id
+        .or_else(|| tokenizer.token_to_id(EOS_TOKEN));
     let prompt = args.prompt.as_ref().map_or(DEFAULT_PROMPT, |p| p.as_str());
     let mut tokens = tokenizer
         .encode(prompt, true)
@@ -160,7 +169,7 @@ fn main() -> Result<()> {
 
     println!("starting the inference loop");
     print!("{prompt}");
-    let mut logits_processor = LogitsProcessor::new(args.seed, args.temperature, args.top_p);
+    let mut logits_processor = LogitsProcessor::new(args.seed, Some(args.temperature), args.top_p);
     let start_gen = std::time::Instant::now();
     let mut index_pos = 0;
     let mut token_generated = 0;

candle-examples/examples/mamba/main.rs

@@ -54,6 +54,7 @@ impl TextGeneration {
     fn run(&mut self, prompt: &str, sample_len: usize) -> Result<()> {
         use std::io::Write;
         self.tokenizer.clear();
+        let dtype = self.model.dtype();
         let mut tokens = self
             .tokenizer
             .tokenizer()
@@ -66,7 +67,7 @@ impl TextGeneration {
             Some(token) => token,
             None => anyhow::bail!("cannot find the </s> token"),
         };
-        let mut state = State::new(1, &self.config, &self.device)?;
+        let mut state = State::new(1, &self.config, dtype, &self.device)?;
         let mut next_logits = None;
         for &t in tokens.iter() {
             let input = Tensor::new(&[t], &self.device)?;
@@ -84,7 +85,7 @@ impl TextGeneration {
                 Some(logits) => logits,
                 None => anyhow::bail!("cannot work on an empty prompt"),
             };
-            let logits = logits.squeeze(0)?.to_dtype(DType::F32)?;
+            let logits = logits.squeeze(0)?.to_dtype(dtype)?;
             let logits = if self.repeat_penalty == 1. {
                 logits
             } else {
@@ -210,6 +211,9 @@ struct Args {
     #[arg(long)]
     config_file: Option<String>,
 
+    #[arg(long, default_value = "f32")]
+    dtype: String,
+
     /// Penalty to be applied for repeating tokens, 1. means no penalty.
     #[arg(long, default_value_t = 1.1)]
     repeat_penalty: f32,
@@ -220,6 +224,7 @@ struct Args {
 }
 
 fn main() -> Result<()> {
+    use std::str::FromStr;
     use tracing_chrome::ChromeLayerBuilder;
     use tracing_subscriber::prelude::*;
 
@@ -279,7 +284,8 @@ fn main() -> Result<()> {
     let start = std::time::Instant::now();
     let config: Config = serde_json::from_slice(&std::fs::read(config_filename)?)?;
     let device = candle_examples::device(args.cpu)?;
-    let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, DType::F32, &device)? };
+    let dtype = DType::from_str(&args.dtype)?;
+    let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
     let model = Model::new(&config, vb.pp("backbone"))?;
     println!("loaded the model in {:?}", start.elapsed());
 

candle-examples/examples/mistral/main.rs

@@ -13,7 +13,7 @@ use candle_transformers::models::quantized_mistral::Model as QMistral;
 use candle::{DType, Device, Tensor};
 use candle_examples::token_output_stream::TokenOutputStream;
 use candle_nn::VarBuilder;
-use candle_transformers::generation::LogitsProcessor;
+use candle_transformers::generation::{LogitsProcessor, Sampling};
 use hf_hub::{api::sync::Api, Repo, RepoType};
 use tokenizers::Tokenizer;
 
@@ -39,11 +39,26 @@ impl TextGeneration {
         seed: u64,
         temp: Option<f64>,
         top_p: Option<f64>,
+        top_k: Option<usize>,
         repeat_penalty: f32,
         repeat_last_n: usize,
         device: &Device,
     ) -> Self {
-        let logits_processor = LogitsProcessor::new(seed, temp, top_p);
+        let logits_processor = {
+            let temperature = temp.unwrap_or(0.);
+            let sampling = if temperature <= 0. {
+                Sampling::ArgMax
+            } else {
+                match (top_k, top_p) {
+                    (None, None) => Sampling::All { temperature },
+                    (Some(k), None) => Sampling::TopK { k, temperature },
+                    (None, Some(p)) => Sampling::TopP { p, temperature },
+                    (Some(k), Some(p)) => Sampling::TopKThenTopP { k, p, temperature },
+                }
+            };
+            LogitsProcessor::from_sampling(seed, sampling)
+        };
+
         Self {
             model,
             tokenizer: TokenOutputStream::new(tokenizer),
@@ -122,6 +137,18 @@ impl TextGeneration {
     }
 }
 
+#[derive(Clone, Debug, Copy, PartialEq, Eq, clap::ValueEnum)]
+enum Which {
+    #[value(name = "7b-v0.1")]
+    Mistral7bV01,
+    #[value(name = "7b-v0.2")]
+    Mistral7bV02,
+    #[value(name = "7b-instruct-v0.1")]
+    Mistral7bInstructV01,
+    #[value(name = "7b-instruct-v0.2")]
+    Mistral7bInstructV02,
+}
+
 #[derive(Parser, Debug)]
 #[command(author, version, about, long_about = None)]
 struct Args {
@@ -147,6 +174,10 @@ struct Args {
     #[arg(long)]
     top_p: Option<f64>,
 
+    /// Only sample among the top K samples.
+    #[arg(long)]
+    top_k: Option<usize>,
+
     /// The seed to use when generating random samples.
     #[arg(long, default_value_t = 299792458)]
     seed: u64,
@@ -155,6 +186,10 @@ struct Args {
     #[arg(long, short = 'n', default_value_t = 10000)]
     sample_len: usize,
 
+    /// The model size to use.
+    #[arg(long, default_value = "7b-v0.1")]
+    which: Which,
+
     #[arg(long)]
     model_id: Option<String>,
 
@@ -164,6 +199,9 @@ struct Args {
     #[arg(long)]
     tokenizer_file: Option<String>,
 
+    #[arg(long)]
+    config_file: Option<String>,
+
     #[arg(long)]
     weight_files: Option<String>,
 
@@ -177,6 +215,10 @@ struct Args {
     /// The context size to consider for the repeat penalty.
     #[arg(long, default_value_t = 64)]
     repeat_last_n: usize,
+
+    /// Use the slower dmmv cuda kernel.
+    #[arg(long)]
+    force_dmmv: bool,
 }
 
 fn main() -> Result<()> {
@@ -184,6 +226,9 @@ fn main() -> Result<()> {
     use tracing_subscriber::prelude::*;
 
     let args = Args::parse();
+    #[cfg(feature = "cuda")]
+    candle::quantized::cuda::set_force_dmmv(args.force_dmmv);
+
     let _guard = if args.tracing {
         let (chrome_layer, guard) = ChromeLayerBuilder::new().build();
         tracing_subscriber::registry().with(chrome_layer).init();
@@ -211,9 +256,17 @@ fn main() -> Result<()> {
         Some(model_id) => model_id,
         None => {
             if args.quantized {
+                if args.which != Which::Mistral7bV01 {
+                    anyhow::bail!("only 7b-v0.1 is available as a quantized model for now")
+                }
                 "lmz/candle-mistral".to_string()
             } else {
-                "mistralai/Mistral-7B-v0.1".to_string()
+                match args.which {
+                    Which::Mistral7bV01 => "mistralai/Mistral-7B-v0.1".to_string(),
+                    Which::Mistral7bV02 => "mistralai/Mistral-7B-v0.2".to_string(),
+                    Which::Mistral7bInstructV01 => "mistralai/Mistral-7B-Instruct-v0.1".to_string(),
+                    Which::Mistral7bInstructV02 => "mistralai/Mistral-7B-Instruct-v0.2".to_string(),
+                }
             }
         }
     };
@@ -243,7 +296,17 @@ fn main() -> Result<()> {
     let tokenizer = Tokenizer::from_file(tokenizer_filename).map_err(E::msg)?;
 
     let start = std::time::Instant::now();
-    let config = Config::config_7b_v0_1(args.use_flash_attn);
+    let config = match args.config_file {
+        Some(config_file) => serde_json::from_slice(&std::fs::read(config_file)?)?,
+        None => {
+            if args.quantized {
+                Config::config_7b_v0_1(args.use_flash_attn)
+            } else {
+                let config_file = repo.get("config.json")?;
+                serde_json::from_slice(&std::fs::read(config_file)?)?
+            }
+        }
+    };
     let device = candle_examples::device(args.cpu)?;
     let (model, device) = if args.quantized {
         let filename = &filenames[0];
@@ -270,6 +333,7 @@ fn main() -> Result<()> {
         args.seed,
         args.temperature,
         args.top_p,
+        args.top_k,
         args.repeat_penalty,
         args.repeat_last_n,
         &device,

candle-examples/examples/moondream/README.md

@@ -0,0 +1,26 @@
+# candle-moondream
+
+[Moondream](https://github.com/vikhyat/moondream) is a computer-vision model can answer real-world questions about images. It's tiny by today's models, with only 1.6B parameters. That enables it to run on a variety of devices, including mobile phones and edge devices.
+
+## Running some examples
+First download an example image
+```bash
+$ wget https://raw.githubusercontent.com/vikhyat/moondream/main/assets/demo-1.jpg
+```
+
+<img src="https://raw.githubusercontent.com/vikhyat/moondream/main/assets/demo-1.jpg" width="200">
+
+Now you can run Moondream from the `candle-examples` crate:
+```bash
+$ cargo run --example moondream --release -- --prompt "What is the girl eating?" --image "./demo-1.jpg"
+
+avavx: false, neon: true, simd128: false, f16c: false
+temp: 0.00 repeat-penalty: 1.00 repeat-last-n: 64
+retrieved the files in 3.395583ms
+Running on CPU, to run on GPU(metal), build this example with `--features metal`
+loaded the model in 5.485493792s
+loaded and encoded the image Tensor[dims 3, 378, 378; f32] in 4.801396417s
+starting the inference loop
+ The girl is eating a hamburger.<
+9 tokens generated (0.68 token/s)
+```

candle-examples/examples/moondream/main.rs

@@ -0,0 +1,343 @@
+#[cfg(feature = "mkl")]
+extern crate intel_mkl_src;
+
+#[cfg(feature = "accelerate")]
+extern crate accelerate_src;
+
+use anyhow::{Error as E, Result};
+use clap::Parser;
+
+use candle::{DType, Device, Tensor};
+use candle_nn::VarBuilder;
+use candle_transformers::{
+    generation::LogitsProcessor,
+    models::{moondream, quantized_moondream},
+};
+use tokenizers::Tokenizer;
+
+enum Model {
+    Moondream(moondream::Model),
+    Quantized(quantized_moondream::Model),
+}
+
+struct TextGeneration {
+    model: Model,
+    device: Device,
+    tokenizer: Tokenizer,
+    logits_processor: LogitsProcessor,
+    repeat_penalty: f32,
+    repeat_last_n: usize,
+    verbose_prompt: bool,
+}
+
+impl TextGeneration {
+    #[allow(clippy::too_many_arguments)]
+    fn new(
+        model: Model,
+        tokenizer: Tokenizer,
+        seed: u64,
+        temp: Option<f64>,
+        top_p: Option<f64>,
+        repeat_penalty: f32,
+        repeat_last_n: usize,
+        verbose_prompt: bool,
+        device: &Device,
+    ) -> Self {
+        let logits_processor = LogitsProcessor::new(seed, temp, top_p);
+        Self {
+            model,
+            tokenizer,
+            logits_processor,
+            repeat_penalty,
+            repeat_last_n,
+            verbose_prompt,
+            device: device.clone(),
+        }
+    }
+
+    fn run(&mut self, prompt: &str, image_embeds: &Tensor, sample_len: usize) -> Result<()> {
+        use std::io::Write;
+        println!("starting the inference loop");
+        let tokens = self.tokenizer.encode(prompt, true).map_err(E::msg)?;
+        if tokens.is_empty() {
+            anyhow::bail!("Empty prompts are not supported in the Moondream model.")
+        }
+        if self.verbose_prompt {
+            for (token, id) in tokens.get_tokens().iter().zip(tokens.get_ids().iter()) {
+                let token = token.replace('▁', " ").replace("<0x0A>", "\n");
+                println!("{id:7} -> '{token}'");
+            }
+        }
+
+        let mut tokens = tokens.get_ids().to_vec();
+        let mut generated_tokens = 0usize;
+
+        // Moondream tokenizer bos_token and eos_token is "<|endoftext|>"
+        // https://huggingface.co/vikhyatk/moondream2/blob/main/special_tokens_map.json
+        let special_token = match self.tokenizer.get_vocab(true).get("<|endoftext|>") {
+            Some(token) => *token,
+            None => anyhow::bail!("cannot find the special token"),
+        };
+        let (bos_token, eos_token) = (special_token, special_token);
+
+        let start_gen = std::time::Instant::now();
+        let mut load_t = std::time::Duration::from_secs_f64(0f64);
+        for index in 0..sample_len {
+            let context_size = if index > 0 { 1 } else { tokens.len() };
+            let ctxt = &tokens[tokens.len().saturating_sub(context_size)..];
+            let input = Tensor::new(ctxt, &self.device)?.unsqueeze(0)?;
+            let logits = if index > 0 {
+                match self.model {
+                    Model::Moondream(ref mut model) => model.text_model.forward(&input)?,
+                    Model::Quantized(ref mut model) => model.text_model.forward(&input)?,
+                }
+            } else {
+                let bos_token = Tensor::new(&[bos_token], &self.device)?.unsqueeze(0)?;
+                let logits = match self.model {
+                    Model::Moondream(ref mut model) => {
+                        model
+                            .text_model
+                            .forward_with_img(&bos_token, &input, image_embeds)?
+                    }
+                    Model::Quantized(ref mut model) => {
+                        model
+                            .text_model
+                            .forward_with_img(&bos_token, &input, image_embeds)?
+                    }
+                };
+                load_t = start_gen.elapsed();
+                println!("load_t: {:?}", load_t);
+                logits
+            };
+            let logits = logits.squeeze(0)?.to_dtype(DType::F32)?;
+            let logits = if self.repeat_penalty == 1. {
+                logits
+            } else {
+                let start_at = tokens.len().saturating_sub(self.repeat_last_n);
+                candle_transformers::utils::apply_repeat_penalty(
+                    &logits,
+                    self.repeat_penalty,
+                    &tokens[start_at..],
+                )?
+            };
+            let next_token = self.logits_processor.sample(&logits)?;
+            tokens.push(next_token);
+            generated_tokens += 1;
+            if next_token == eos_token || tokens.ends_with(&[27, 10619, 29] /* <END> */) {
+                break;
+            }
+            let token = self.tokenizer.decode(&[next_token], true).map_err(E::msg)?;
+            print!("{token}");
+            std::io::stdout().flush()?;
+        }
+
+        let dt = start_gen.elapsed() - load_t;
+        println!(
+            "\ngenerated in {} seconds\n{generated_tokens} tokens generated ({:.2} token/s)",
+            dt.as_secs_f64(),
+            (generated_tokens - 1) as f64 / dt.as_secs_f64()
+        );
+
+        Ok(())
+    }
+}
+
+#[derive(Parser)]
+struct Args {
+    /// Run on CPU rather than on GPU.
+    #[arg(long)]
+    cpu: bool,
+
+    /// Enable tracing (generates a trace-timestamp.json file).
+    #[arg(long)]
+    tracing: bool,
+
+    /// Display the token for the specified prompt.
+    #[arg(long)]
+    verbose_prompt: bool,
+
+    #[arg(long)]
+    prompt: String,
+
+    #[arg(long)]
+    image: String,
+
+    /// The temperature used to generate samples.
+    #[arg(long)]
+    temperature: Option<f64>,
+
+    /// Nucleus sampling probability cutoff.
+    #[arg(long)]
+    top_p: Option<f64>,
+
+    /// The seed to use when generating random samples.
+    #[arg(long, default_value_t = 0)]
+    seed: u64,
+
+    #[arg(long, default_value_t = 5000)]
+    sample_len: usize,
+
+    /// Penalty to be applied for repeating tokens, 1. means no penalty.
+    #[arg(long, default_value_t = 1.0)]
+    repeat_penalty: f32,
+
+    /// The context size to consider for the repeat penalty.
+    #[arg(long, default_value_t = 64)]
+    repeat_last_n: usize,
+
+    #[arg(long)]
+    model_id: Option<String>,
+
+    #[arg(long, default_value = "main")]
+    revision: String,
+
+    #[arg(long)]
+    quantized: bool,
+
+    /// Use f16 precision for all the computations rather than f32.
+    #[arg(long)]
+    f16: bool,
+
+    #[arg(long)]
+    model_file: Option<String>,
+
+    #[arg(long)]
+    tokenizer_file: Option<String>,
+}
+
+/// Loads an image from disk using the image crate, this returns a tensor with shape
+/// (3, 378, 378).
+pub fn load_image<P: AsRef<std::path::Path>>(p: P) -> candle::Result<Tensor> {
+    let img = image::io::Reader::open(p)?
+        .decode()
+        .map_err(candle::Error::wrap)?
+        .resize_to_fill(378, 378, image::imageops::FilterType::Triangle); // Adjusted to 378x378
+    let img = img.to_rgb8();
+    let data = img.into_raw();
+    let data = Tensor::from_vec(data, (378, 378, 3), &Device::Cpu)?.permute((2, 0, 1))?;
+    let mean = Tensor::new(&[0.5f32, 0.5, 0.5], &Device::Cpu)?.reshape((3, 1, 1))?;
+    let std = Tensor::new(&[0.5f32, 0.5, 0.5], &Device::Cpu)?.reshape((3, 1, 1))?;
+    (data.to_dtype(candle::DType::F32)? / 255.)?
+        .broadcast_sub(&mean)?
+        .broadcast_div(&std)
+}
+
+#[tokio::main]
+async fn main() -> anyhow::Result<()> {
+    use tracing_chrome::ChromeLayerBuilder;
+    use tracing_subscriber::prelude::*;
+
+    let args = Args::parse();
+
+    let _guard = if args.tracing {
+        let (chrome_layer, guard) = ChromeLayerBuilder::new().build();
+        tracing_subscriber::registry().with(chrome_layer).init();
+        Some(guard)
+    } else {
+        None
+    };
+    println!(
+        "avx: {}, neon: {}, simd128: {}, f16c: {}",
+        candle::utils::with_avx(),
+        candle::utils::with_neon(),
+        candle::utils::with_simd128(),
+        candle::utils::with_f16c()
+    );
+    println!(
+        "temp: {:.2} repeat-penalty: {:.2} repeat-last-n: {}",
+        args.temperature.unwrap_or(0.),
+        args.repeat_penalty,
+        args.repeat_last_n
+    );
+
+    let start = std::time::Instant::now();
+    let api = hf_hub::api::tokio::Api::new()?;
+    let model_id = match args.model_id {
+        Some(model_id) => model_id.to_string(),
+        None => {
+            if args.quantized {
+                "santiagomed/candle-moondream".to_string()
+            } else {
+                "vikhyatk/moondream2".to_string()
+            }
+        }
+    };
+    let repo = api.repo(hf_hub::Repo::with_revision(
+        model_id,
+        hf_hub::RepoType::Model,
+        args.revision,
+    ));
+    let model_file = match args.model_file {
+        Some(m) => m.into(),
+        None => {
+            if args.quantized {
+                repo.get("model-q4_0.gguf").await?
+            } else {
+                repo.get("model.safetensors").await?
+            }
+        }
+    };
+    let tokenizer = match args.tokenizer_file {
+        Some(m) => m.into(),
+        None => repo.get("tokenizer.json").await?,
+    };
+    println!("retrieved the files in {:?}", start.elapsed());
+    let tokenizer = Tokenizer::from_file(tokenizer).map_err(E::msg)?;
+
+    let start = std::time::Instant::now();
+    let device = candle_examples::device(args.cpu)?;
+    let config = moondream::Config::v2();
+    let dtype = if args.quantized {
+        if args.f16 {
+            anyhow::bail!("Quantized model does not support f16");
+        }
+        DType::F32
+    } else if device.is_cuda() || args.f16 {
+        DType::F16
+    } else {
+        DType::F32
+    };
+    let model = if args.quantized {
+        let vb = candle_transformers::quantized_var_builder::VarBuilder::from_gguf(
+            &model_file,
+            &device,
+        )?;
+        let model = quantized_moondream::Model::new(&config, vb)?;
+        Model::Quantized(model)
+    } else {
+        let vb = unsafe { VarBuilder::from_mmaped_safetensors(&[model_file], dtype, &device)? };
+        let model = moondream::Model::new(&config, vb)?;
+        Model::Moondream(model)
+    };
+    println!("loaded the model in {:?}", start.elapsed());
+
+    let start = std::time::Instant::now();
+    let image = load_image(args.image)?
+        .to_device(&device)?
+        .to_dtype(dtype)?;
+    let image_embeds = image.unsqueeze(0)?;
+    let image_embeds = match model {
+        Model::Moondream(ref m) => image_embeds.apply(m.vision_encoder())?,
+        Model::Quantized(ref m) => image_embeds.apply(m.vision_encoder())?,
+    };
+    println!(
+        "loaded and encoded the image {image:?} in {:?}",
+        start.elapsed()
+    );
+
+    let prompt = format!("\n\nQuestion: {0}\n\nAnswer:", args.prompt);
+    let mut pipeline = TextGeneration::new(
+        model,
+        tokenizer,
+        args.seed,
+        args.temperature,
+        args.top_p,
+        args.repeat_penalty,
+        args.repeat_last_n,
+        args.verbose_prompt,
+        &device,
+    );
+    pipeline.run(&prompt, &image_embeds, args.sample_len)?;
+
+    Ok(())
+}

candle-examples/examples/quantized-t5/README.md

@@ -17,7 +17,7 @@ generate quantized weight files from the original safetensors file by using the
 `tensor-tools` command line utility via:
 
 ```bash
-$ cargo run --example tensor-tools --release -- quantize --quantization q6k PATH/TO/T5/model.safetensors /tmp/model.gguf
+$ cargo run --bin tensor-tools --release -- quantize --quantization q6k PATH/TO/T5/model.safetensors /tmp/model.gguf
 ```
 
 ## Using custom models

candle-examples/examples/quantized/main.rs

@@ -10,7 +10,7 @@ use tokenizers::Tokenizer;
 
 use candle::quantized::{ggml_file, gguf_file};
 use candle::Tensor;
-use candle_transformers::generation::LogitsProcessor;
+use candle_transformers::generation::{LogitsProcessor, Sampling};
 
 use candle_examples::token_output_stream::TokenOutputStream;
 use candle_transformers::models::quantized_llama as model;
@@ -67,6 +67,8 @@ enum Which {
     Mixtral,
     #[value(name = "mixtral-instruct")]
     MixtralInstruct,
+    #[value(name = "llama3-8b")]
+    L8b,
 }
 
 impl Which {
@@ -82,7 +84,8 @@ impl Which {
             | Self::L13bCode
             | Self::L34bCode
             | Self::Leo7b
-            | Self::Leo13b => false,
+            | Self::Leo13b
+            | Self::L8b => false,
             // Zephyr and OpenChat are fine tuned versions of mistral and should be treated in the
             // same way. Starling is a fine tuned version of OpenChat.
             Self::OpenChat35
@@ -116,7 +119,8 @@ impl Which {
             | Self::Mistral7bInstruct
             | Self::Mistral7bInstructV02
             | Self::OpenChat35
-            | Self::Starling7bAlpha => false,
+            | Self::Starling7bAlpha
+            | Self::L8b => false,
             Self::Zephyr7bAlpha | Self::Zephyr7bBeta => true,
         }
     }
@@ -140,7 +144,8 @@ impl Which {
             | Self::Mistral7bInstruct
             | Self::Mistral7bInstructV02
             | Self::Zephyr7bAlpha
-            | Self::Zephyr7bBeta => false,
+            | Self::Zephyr7bBeta
+            | Self::L8b => false,
             Self::OpenChat35 | Self::Starling7bAlpha => true,
         }
     }
@@ -167,6 +172,7 @@ impl Which {
             | Which::Zephyr7bBeta => "mistralai/Mistral-7B-v0.1",
             Which::OpenChat35 => "openchat/openchat_3.5",
             Which::Starling7bAlpha => "berkeley-nest/Starling-LM-7B-alpha",
+            Self::L8b => "meta-llama/Meta-Llama-3-8B",
         }
     }
 }
@@ -200,6 +206,10 @@ struct Args {
     #[arg(long)]
     top_p: Option<f64>,
 
+    /// Only sample among the top K samples.
+    #[arg(long)]
+    top_k: Option<usize>,
+
     /// The seed to use when generating random samples.
     #[arg(long, default_value_t = 299792458)]
     seed: u64,
@@ -235,6 +245,10 @@ struct Args {
     /// Group-Query Attention, use 8 for the 70B version of LLaMAv2.
     #[arg(long)]
     gqa: Option<usize>,
+
+    /// Use the slower dmmv cuda kernel.
+    #[arg(long)]
+    force_dmmv: bool,
 }
 
 impl Args {
@@ -314,6 +328,11 @@ impl Args {
                         "TheBloke/Starling-LM-7B-alpha-GGUF",
                         "starling-lm-7b-alpha.Q4_K_M.gguf",
                     ),
+                    // TODO: swap to TheBloke model when available
+                    Which::L8b => (
+                        "QuantFactory/Meta-Llama-3-8B-GGUF",
+                        "Meta-Llama-3-8B.Q4_K_S.gguf",
+                    ),
                 };
                 let api = hf_hub::api::sync::Api::new()?;
                 let api = api.model(repo.to_string());
@@ -341,11 +360,10 @@ fn main() -> anyhow::Result<()> {
     use tracing_subscriber::prelude::*;
 
     let args = Args::parse();
-    let temperature = if args.temperature == 0. {
-        None
-    } else {
-        Some(args.temperature)
-    };
+
+    #[cfg(feature = "cuda")]
+    candle::quantized::cuda::set_force_dmmv(args.force_dmmv);
+
     let _guard = if args.tracing {
         let (chrome_layer, guard) = ChromeLayerBuilder::new().build();
         tracing_subscriber::registry().with(chrome_layer).init();
@@ -413,7 +431,8 @@ fn main() -> anyhow::Result<()> {
                 | Which::L13bCode
                 | Which::L34bCode
                 | Which::Leo7b
-                | Which::Leo13b => 1,
+                | Which::Leo13b
+                | Which::L8b => 1,
                 Which::Mixtral
                 | Which::MixtralInstruct
                 | Which::Mistral7b
@@ -492,7 +511,20 @@ fn main() -> anyhow::Result<()> {
             prompt_tokens
         };
         let mut all_tokens = vec![];
-        let mut logits_processor = LogitsProcessor::new(args.seed, temperature, args.top_p);
+        let mut logits_processor = {
+            let temperature = args.temperature;
+            let sampling = if temperature <= 0. {
+                Sampling::ArgMax
+            } else {
+                match (args.top_k, args.top_p) {
+                    (None, None) => Sampling::All { temperature },
+                    (Some(k), None) => Sampling::TopK { k, temperature },
+                    (None, Some(p)) => Sampling::TopP { p, temperature },
+                    (Some(k), Some(p)) => Sampling::TopKThenTopP { k, p, temperature },
+                }
+            };
+            LogitsProcessor::from_sampling(args.seed, sampling)
+        };
 
         let start_prompt_processing = std::time::Instant::now();
         let mut next_token = if !args.split_prompt {
@@ -517,11 +549,14 @@ fn main() -> anyhow::Result<()> {
             std::io::stdout().flush()?;
         }
 
-        let eos_token = if args.which.is_open_chat() {
-            "<|end_of_turn|>"
-        } else {
-            "</s>"
+        let eos_token = match args.which {
+            Which::L8b => "<|end_of_text|>",
+            _ => match args.which.is_open_chat() {
+                true => "<|end_of_turn|>",
+                false => "</s>",
+            },
         };
+
         let eos_token = *tos.tokenizer().get_vocab(true).get(eos_token).unwrap();
         let start_post_prompt = std::time::Instant::now();
         let mut sampled = 0;

candle-examples/examples/qwen/README.md

@@ -0,0 +1,27 @@
+# candle-qwen: large language model series from Alibaba Cloud
+
+Qwen 1.5 is a series of large language models that provide strong performances
+on English and Chinese.
+
+- [Blog post](https://qwenlm.github.io/blog/qwen1.5/) introducing Qwen1.5.
+- [Model card](https://huggingface.co/Qwen/Qwen1.5-0.5B) on the HuggingFace Hub.
+- [Blog post](https://qwenlm.github.io/blog/qwen-moe/) for the
+  mixture-of-experts (MoE) variant.
+
+## Running the example
+
+```bash
+$ cargo run --example qwen --release  -- --prompt "Hello there "
+```
+
+Various model sizes are available via the `--model` argument, including the MoE
+variant.
+
+```bash
+$ cargo run --example qwen --release  -- --model moe-a2.7b --prompt 'def print_prime(n: int): '
+def print_prime(n: int):  # n is the number of primes to be printed
+    for i in range(2, n + 1):
+        if all(i % j != 0 for j in range(2, i)):
+            print(i)
+```
+

candle-examples/examples/qwen/main.rs

@@ -7,7 +7,8 @@ extern crate accelerate_src;
 use anyhow::{Error as E, Result};
 use clap::Parser;
 
-use candle_transformers::models::qwen2::{Config, Model};
+use candle_transformers::models::qwen2::{Config as ConfigBase, Model as ModelBase};
+use candle_transformers::models::qwen2_moe::{Config as ConfigMoe, Model as ModelMoe};
 
 use candle::{DType, Device, Tensor};
 use candle_examples::token_output_stream::TokenOutputStream;
@@ -16,6 +17,20 @@ use candle_transformers::generation::LogitsProcessor;
 use hf_hub::{api::sync::Api, Repo, RepoType};
 use tokenizers::Tokenizer;
 
+enum Model {
+    Base(ModelBase),
+    Moe(ModelMoe),
+}
+
+impl Model {
+    fn forward(&mut self, xs: &Tensor, s: usize) -> candle::Result<Tensor> {
+        match self {
+            Self::Moe(ref mut m) => m.forward(xs, s),
+            Self::Base(ref mut m) => m.forward(xs, s),
+        }
+    }
+}
+
 struct TextGeneration {
     model: Model,
     device: Device,
@@ -127,6 +142,8 @@ enum WhichModel {
     W14b,
     #[value(name = "72b")]
     W72b,
+    #[value(name = "moe-a2.7b")]
+    MoeA27b,
 }
 
 #[derive(Parser, Debug)]
@@ -224,6 +241,7 @@ fn main() -> Result<()> {
                 WhichModel::W7b => "7B",
                 WhichModel::W14b => "14B",
                 WhichModel::W72b => "72B",
+                WhichModel::MoeA27b => "MoE-A2.7B",
             };
             format!("Qwen/Qwen1.5-{size}")
         }
@@ -244,7 +262,11 @@ fn main() -> Result<()> {
             .collect::<Vec<_>>(),
         None => match args.model {
             WhichModel::W0_5b | WhichModel::W1_8b => vec![repo.get("model.safetensors")?],
-            WhichModel::W4b | WhichModel::W7b | WhichModel::W14b | WhichModel::W72b => {
+            WhichModel::W4b
+            | WhichModel::W7b
+            | WhichModel::W14b
+            | WhichModel::W72b
+            | WhichModel::MoeA27b => {
                 candle_examples::hub_load_safetensors(&repo, "model.safetensors.index.json")?
             }
         },
@@ -254,7 +276,6 @@ fn main() -> Result<()> {
 
     let start = std::time::Instant::now();
     let config_file = repo.get("config.json")?;
-    let config: Config = serde_json::from_slice(&std::fs::read(config_file)?)?;
     let device = candle_examples::device(args.cpu)?;
     let dtype = if device.is_cuda() {
         DType::BF16
@@ -262,7 +283,16 @@ fn main() -> Result<()> {
         DType::F32
     };
     let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
-    let model = Model::new(&config, vb)?;
+    let model = match args.model {
+        WhichModel::MoeA27b => {
+            let config: ConfigMoe = serde_json::from_slice(&std::fs::read(config_file)?)?;
+            Model::Moe(ModelMoe::new(&config, vb)?)
+        }
+        _ => {
+            let config: ConfigBase = serde_json::from_slice(&std::fs::read(config_file)?)?;
+            Model::Base(ModelBase::new(&config, vb)?)
+        }
+    };
 
     println!("loaded the model in {:?}", start.elapsed());
 

candle-examples/examples/recurrent-gemma/README.md

@@ -0,0 +1,9 @@
+# candle-recurrent-gemma
+
+This model card corresponds to the 2B base version of the RecurrentGemma model
+[huggingface model card](https://huggingface.co/google/recurrentgemma-2b).
+
+```bash
+cargo run --features cuda -r --example recurrent-gemma -- \
+    --prompt "Write me a poem about Machine Learning."  
+```

candle-examples/examples/recurrent-gemma/main.rs

@@ -0,0 +1,321 @@
+#[cfg(feature = "mkl")]
+extern crate intel_mkl_src;
+
+#[cfg(feature = "accelerate")]
+extern crate accelerate_src;
+
+use anyhow::{Error as E, Result};
+use clap::Parser;
+
+use candle_transformers::models::quantized_recurrent_gemma::Model as QModel;
+use candle_transformers::models::recurrent_gemma::{Config, Model as BModel};
+
+use candle::{DType, Device, Tensor};
+use candle_examples::token_output_stream::TokenOutputStream;
+use candle_nn::VarBuilder;
+use candle_transformers::generation::LogitsProcessor;
+use hf_hub::{api::sync::Api, Repo, RepoType};
+use tokenizers::Tokenizer;
+
+enum Model {
+    B(BModel),
+    Q(QModel),
+}
+
+impl Model {
+    fn forward(&mut self, xs: &Tensor, pos: usize) -> candle::Result<Tensor> {
+        match self {
+            Self::B(m) => m.forward(xs, pos),
+            Self::Q(m) => m.forward(xs, pos),
+        }
+    }
+}
+
+#[derive(Clone, Debug, Copy, PartialEq, Eq, clap::ValueEnum)]
+enum Which {
+    #[value(name = "2b")]
+    Base2B,
+    #[value(name = "2b-it")]
+    Instruct2B,
+}
+
+struct TextGeneration {
+    model: Model,
+    device: Device,
+    tokenizer: TokenOutputStream,
+    logits_processor: LogitsProcessor,
+    repeat_penalty: f32,
+    repeat_last_n: usize,
+}
+
+impl TextGeneration {
+    #[allow(clippy::too_many_arguments)]
+    fn new(
+        model: Model,
+        tokenizer: Tokenizer,
+        seed: u64,
+        temp: Option<f64>,
+        top_p: Option<f64>,
+        top_k: usize,
+        repeat_penalty: f32,
+        repeat_last_n: usize,
+        device: &Device,
+    ) -> Self {
+        let sampling = match temp {
+            None => candle_transformers::generation::Sampling::ArgMax,
+            Some(temperature) => match top_p {
+                None => candle_transformers::generation::Sampling::TopK {
+                    temperature,
+                    k: top_k,
+                },
+                Some(top_p) => candle_transformers::generation::Sampling::TopKThenTopP {
+                    temperature,
+                    k: top_k,
+                    p: top_p,
+                },
+            },
+        };
+        let logits_processor = LogitsProcessor::from_sampling(seed, sampling);
+        Self {
+            model,
+            tokenizer: TokenOutputStream::new(tokenizer),
+            logits_processor,
+            repeat_penalty,
+            repeat_last_n,
+            device: device.clone(),
+        }
+    }
+
+    fn run(&mut self, prompt: &str, sample_len: usize) -> Result<()> {
+        use std::io::Write;
+        self.tokenizer.clear();
+        let mut tokens = self
+            .tokenizer
+            .tokenizer()
+            .encode(prompt, true)
+            .map_err(E::msg)?
+            .get_ids()
+            .to_vec();
+        for &t in tokens.iter() {
+            if let Some(t) = self.tokenizer.next_token(t)? {
+                print!("{t}")
+            }
+        }
+        std::io::stdout().flush()?;
+
+        let mut generated_tokens = 0usize;
+        let eos_token = match self.tokenizer.get_token("<eos>") {
+            Some(token) => token,
+            None => anyhow::bail!("cannot find the <eos> token"),
+        };
+        let start_gen = std::time::Instant::now();
+        for index in 0..sample_len {
+            let context_size = if index > 0 { 1 } else { tokens.len() };
+            let start_pos = tokens.len().saturating_sub(context_size);
+            let ctxt = &tokens[start_pos..];
+            let input = Tensor::new(ctxt, &self.device)?.unsqueeze(0)?;
+            let logits = self.model.forward(&input, start_pos)?;
+            let logits = logits.squeeze(0)?.squeeze(0)?.to_dtype(DType::F32)?;
+            let logits = if self.repeat_penalty == 1. {
+                logits
+            } else {
+                let start_at = tokens.len().saturating_sub(self.repeat_last_n);
+                candle_transformers::utils::apply_repeat_penalty(
+                    &logits,
+                    self.repeat_penalty,
+                    &tokens[start_at..],
+                )?
+            };
+
+            let next_token = self.logits_processor.sample(&logits)?;
+            tokens.push(next_token);
+            generated_tokens += 1;
+            if next_token == eos_token {
+                break;
+            }
+            if let Some(t) = self.tokenizer.next_token(next_token)? {
+                print!("{t}");
+                std::io::stdout().flush()?;
+            }
+        }
+        let dt = start_gen.elapsed();
+        if let Some(rest) = self.tokenizer.decode_rest().map_err(E::msg)? {
+            print!("{rest}");
+        }
+        std::io::stdout().flush()?;
+        println!(
+            "\n{generated_tokens} tokens generated ({:.2} token/s)",
+            generated_tokens as f64 / dt.as_secs_f64(),
+        );
+        Ok(())
+    }
+}
+
+#[derive(Parser, Debug)]
+#[command(author, version, about, long_about = None)]
+struct Args {
+    /// Run on CPU rather than on GPU.
+    #[arg(long)]
+    cpu: bool,
+
+    /// Enable tracing (generates a trace-timestamp.json file).
+    #[arg(long)]
+    tracing: bool,
+
+    #[arg(long)]
+    prompt: String,
+
+    /// The temperature used to generate samples.
+    #[arg(long)]
+    temperature: Option<f64>,
+
+    /// Nucleus sampling probability cutoff.
+    #[arg(long)]
+    top_p: Option<f64>,
+
+    #[arg(long, default_value_t = 250)]
+    top_k: usize,
+
+    /// The seed to use when generating random samples.
+    #[arg(long, default_value_t = 299792458)]
+    seed: u64,
+
+    /// The length of the sample to generate (in tokens).
+    #[arg(long, short = 'n', default_value_t = 8000)]
+    sample_len: usize,
+
+    #[arg(long)]
+    model_id: Option<String>,
+
+    #[arg(long, default_value = "main")]
+    revision: String,
+
+    #[arg(long)]
+    tokenizer_file: Option<String>,
+
+    #[arg(long)]
+    config_file: Option<String>,
+
+    #[arg(long)]
+    weight_files: Option<String>,
+
+    /// Penalty to be applied for repeating tokens, 1. means no penalty.
+    #[arg(long, default_value_t = 1.1)]
+    repeat_penalty: f32,
+
+    /// The context size to consider for the repeat penalty.
+    #[arg(long, default_value_t = 64)]
+    repeat_last_n: usize,
+
+    /// The model to use.
+    #[arg(long, default_value = "2b")]
+    which: Which,
+
+    #[arg(long)]
+    quantized: bool,
+}
+
+fn main() -> Result<()> {
+    use tracing_chrome::ChromeLayerBuilder;
+    use tracing_subscriber::prelude::*;
+
+    let args = Args::parse();
+    let _guard = if args.tracing {
+        let (chrome_layer, guard) = ChromeLayerBuilder::new().build();
+        tracing_subscriber::registry().with(chrome_layer).init();
+        Some(guard)
+    } else {
+        None
+    };
+    println!(
+        "avx: {}, neon: {}, simd128: {}, f16c: {}",
+        candle::utils::with_avx(),
+        candle::utils::with_neon(),
+        candle::utils::with_simd128(),
+        candle::utils::with_f16c()
+    );
+    println!(
+        "temp: {:.2} repeat-penalty: {:.2} repeat-last-n: {}",
+        args.temperature.unwrap_or(0.),
+        args.repeat_penalty,
+        args.repeat_last_n
+    );
+
+    let start = std::time::Instant::now();
+    let api = Api::new()?;
+    let model_id = match &args.model_id {
+        Some(model_id) => model_id.to_string(),
+        None => match args.which {
+            Which::Base2B => "google/recurrentgemma-2b".to_string(),
+            Which::Instruct2B => "google/recurrentgemma-2b-it".to_string(),
+        },
+    };
+    let repo = api.repo(Repo::with_revision(
+        model_id,
+        RepoType::Model,
+        args.revision,
+    ));
+    let tokenizer_filename = match args.tokenizer_file {
+        Some(file) => std::path::PathBuf::from(file),
+        None => repo.get("tokenizer.json")?,
+    };
+    let config_filename = match args.config_file {
+        Some(file) => std::path::PathBuf::from(file),
+        None => repo.get("config.json")?,
+    };
+    let filenames = match args.weight_files {
+        Some(files) => files
+            .split(',')
+            .map(std::path::PathBuf::from)
+            .collect::<Vec<_>>(),
+        None => {
+            if args.quantized {
+                let filename = match args.which {
+                    Which::Base2B => "recurrent-gemma-2b-q4k.gguf",
+                    Which::Instruct2B => "recurrent-gemma-7b-q4k.gguf",
+                };
+                let filename = api.model("lmz/candle-gemma".to_string()).get(filename)?;
+                vec![filename]
+            } else {
+                candle_examples::hub_load_safetensors(&repo, "model.safetensors.index.json")?
+            }
+        }
+    };
+    println!("retrieved the files in {:?}", start.elapsed());
+    let tokenizer = Tokenizer::from_file(tokenizer_filename).map_err(E::msg)?;
+    let config: Config = serde_json::from_reader(std::fs::File::open(config_filename)?)?;
+
+    let start = std::time::Instant::now();
+    let device = candle_examples::device(args.cpu)?;
+    let dtype = if device.is_cuda() {
+        DType::BF16
+    } else {
+        DType::F32
+    };
+    let model = if args.quantized {
+        let vb = candle_transformers::quantized_var_builder::VarBuilder::from_gguf(
+            &filenames[0],
+            &device,
+        )?;
+        Model::Q(QModel::new(&config, vb.pp("model"))?)
+    } else {
+        let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
+        Model::B(BModel::new(&config, vb.pp("model"))?)
+    };
+
+    println!("loaded the model in {:?}", start.elapsed());
+
+    let mut pipeline = TextGeneration::new(
+        model,
+        tokenizer,
+        args.seed,
+        args.temperature,
+        args.top_p,
+        args.top_k,
+        args.repeat_penalty,
+        args.repeat_last_n,
+        &device,
+    );
+    pipeline.run(&args.prompt, args.sample_len)?;
+    Ok(())
+}

candle-examples/examples/reinforcement-learning/dqn.rs

@@ -0,0 +1,118 @@
+use std::collections::VecDeque;
+
+use rand::distributions::Uniform;
+use rand::{thread_rng, Rng};
+
+use candle::{DType, Device, Module, Result, Tensor};
+use candle_nn::loss::mse;
+use candle_nn::{linear, seq, Activation, AdamW, Optimizer, VarBuilder, VarMap};
+
+use crate::gym_env::GymEnv;
+
+const DEVICE: Device = Device::Cpu;
+const EPISODES: usize = 200;
+const BATCH_SIZE: usize = 64;
+const GAMMA: f64 = 0.99;
+const LEARNING_RATE: f64 = 0.01;
+
+pub fn run() -> Result<()> {
+    let env = GymEnv::new("CartPole-v1")?;
+
+    // Build the model that predicts the estimated rewards given a specific state.
+    let var_map = VarMap::new();
+    let vb = VarBuilder::from_varmap(&var_map, DType::F32, &DEVICE);
+    let observation_space = *env.observation_space().first().unwrap();
+
+    let model = seq()
+        .add(linear(observation_space, 64, vb.pp("linear_in"))?)
+        .add(Activation::Relu)
+        .add(linear(64, env.action_space(), vb.pp("linear_out"))?);
+
+    let mut optimizer = AdamW::new_lr(var_map.all_vars(), LEARNING_RATE)?;
+
+    // Initialize the model's memory.
+    let mut memory = VecDeque::with_capacity(10000);
+
+    // Start the training loop.
+    let mut state = env.reset(0)?;
+    let mut episode = 0;
+    let mut accumulate_rewards = 0.0;
+    while episode < EPISODES {
+        // Given the current state, predict the estimated rewards, and take the
+        // action that is expected to return the most rewards.
+        let estimated_rewards = model.forward(&state.unsqueeze(0)?)?;
+        let action: u32 = estimated_rewards.squeeze(0)?.argmax(0)?.to_scalar()?;
+
+        // Take that action in the environment, and memorize the outcome:
+        // - the state for which the action was taken
+        // - the action taken
+        // - the new state resulting of taking that action
+        // - the actual rewards of taking that action
+        // - whether the environment reached a terminal state or not (e.g. game over)
+        let step = env.step(action)?;
+        accumulate_rewards += step.reward;
+        memory.push_back((
+            state,
+            action,
+            step.state.clone(),
+            step.reward,
+            step.terminated || step.truncated,
+        ));
+        state = step.state;
+
+        // If there's enough entries in the memory, perform a learning step, where
+        // BATCH_SIZE transitions will be sampled from the memory and will be
+        // fed to the model so that it performs a backward pass.
+        if memory.len() > BATCH_SIZE {
+            // Sample randomly from the memory.
+            let batch = thread_rng()
+                .sample_iter(Uniform::from(0..memory.len()))
+                .take(BATCH_SIZE)
+                .map(|i| memory.get(i).unwrap().clone())
+                .collect::<Vec<_>>();
+
+            // Group all the samples together into tensors with the appropriate shape.
+            let states: Vec<_> = batch.iter().map(|e| e.0.clone()).collect();
+            let states = Tensor::stack(&states, 0)?;
+
+            let actions = batch.iter().map(|e| e.1);
+            let actions = Tensor::from_iter(actions, &DEVICE)?.unsqueeze(1)?;
+
+            let next_states: Vec<_> = batch.iter().map(|e| e.2.clone()).collect();
+            let next_states = Tensor::stack(&next_states, 0)?;
+
+            let rewards = batch.iter().map(|e| e.3 as f32);
+            let rewards = Tensor::from_iter(rewards, &DEVICE)?.unsqueeze(1)?;
+
+            let non_final_mask = batch.iter().map(|e| !e.4 as u8 as f32);
+            let non_final_mask = Tensor::from_iter(non_final_mask, &DEVICE)?.unsqueeze(1)?;
+
+            // Get the estimated rewards for the actions that where taken at each step.
+            let estimated_rewards = model.forward(&states)?;
+            let x = estimated_rewards.gather(&actions, 1)?;
+
+            // Get the maximum expected rewards for the next state, apply them a discount rate
+            // GAMMA and add them to the rewards that were actually gathered on the current state.
+            // If the next state is a terminal state, just omit maximum estimated
+            // rewards for that state.
+            let expected_rewards = model.forward(&next_states)?.detach();
+            let y = expected_rewards.max_keepdim(1)?;
+            let y = (y * GAMMA * non_final_mask + rewards)?;
+
+            // Compare the estimated rewards with the maximum expected rewards and
+            // perform the backward step.
+            let loss = mse(&x, &y)?;
+            optimizer.backward_step(&loss)?;
+        }
+
+        // If we are on a terminal state, reset the environment and log how it went.
+        if step.terminated || step.truncated {
+            episode += 1;
+            println!("Episode {episode} | Rewards {}", accumulate_rewards as i64);
+            state = env.reset(0)?;
+            accumulate_rewards = 0.0;
+        }
+    }
+
+    Ok(())
+}

candle-examples/examples/reinforcement-learning/gym_env.rs

@@ -42,7 +42,7 @@ impl GymEnv {
     /// Creates a new session of the specified OpenAI Gym environment.
     pub fn new(name: &str) -> Result<GymEnv> {
         Python::with_gil(|py| {
-            let gym = py.import("gymnasium")?;
+            let gym = py.import_bound("gymnasium")?;
             let make = gym.getattr("make")?;
             let env = make.call1((name,))?;
             let action_space = env.getattr("action_space")?;
@@ -66,10 +66,10 @@ impl GymEnv {
     /// Resets the environment, returning the observation tensor.
     pub fn reset(&self, seed: u64) -> Result<Tensor> {
         let state: Vec<f32> = Python::with_gil(|py| {
-            let kwargs = PyDict::new(py);
+            let kwargs = PyDict::new_bound(py);
             kwargs.set_item("seed", seed)?;
-            let state = self.env.call_method(py, "reset", (), Some(kwargs))?;
-            state.as_ref(py).get_item(0)?.extract()
+            let state = self.env.call_method_bound(py, "reset", (), Some(&kwargs))?;
+            state.bind(py).get_item(0)?.extract()
         })
         .map_err(w)?;
         Tensor::new(state, &Device::Cpu)
@@ -81,8 +81,10 @@ impl GymEnv {
         action: A,
     ) -> Result<Step<A>> {
         let (state, reward, terminated, truncated) = Python::with_gil(|py| {
-            let step = self.env.call_method(py, "step", (action.clone(),), None)?;
-            let step = step.as_ref(py);
+            let step = self
+                .env
+                .call_method_bound(py, "step", (action.clone(),), None)?;
+            let step = step.bind(py);
             let state: Vec<f32> = step.get_item(0)?.extract()?;
             let reward: f64 = step.get_item(1)?.extract()?;
             let terminated: bool = step.get_item(2)?.extract()?;

candle-examples/examples/reinforcement-learning/main.rs

@@ -13,6 +13,7 @@ mod gym_env;
 mod vec_gym_env;
 
 mod ddpg;
+mod dqn;
 mod policy_gradient;
 
 #[derive(Parser)]
@@ -25,6 +26,7 @@ struct Args {
 enum Command {
     Pg,
     Ddpg,
+    Dqn,
 }
 
 fn main() -> Result<()> {
@@ -32,6 +34,7 @@ fn main() -> Result<()> {
     match args.command {
         Command::Pg => policy_gradient::run()?,
         Command::Ddpg => ddpg::run()?,
+        Command::Dqn => dqn::run()?,
     }
     Ok(())
 }

candle-examples/examples/reinforcement-learning/vec_gym_env.rs

@@ -24,13 +24,13 @@ fn w(res: PyErr) -> candle::Error {
 impl VecGymEnv {
     pub fn new(name: &str, img_dir: Option<&str>, nprocesses: usize) -> Result<VecGymEnv> {
         Python::with_gil(|py| {
-            let sys = py.import("sys")?;
+            let sys = py.import_bound("sys")?;
             let path = sys.getattr("path")?;
             let _ = path.call_method1(
                 "append",
                 ("candle-examples/examples/reinforcement-learning",),
             )?;
-            let gym = py.import("atari_wrappers")?;
+            let gym = py.import_bound("atari_wrappers")?;
             let make = gym.getattr("make")?;
             let env = make.call1((name, img_dir, nprocesses))?;
             let action_space = env.getattr("action_space")?;
@@ -60,10 +60,10 @@ impl VecGymEnv {
 
     pub fn step(&self, action: Vec<usize>) -> Result<Step> {
         let (obs, reward, is_done) = Python::with_gil(|py| {
-            let step = self.env.call_method(py, "step", (action,), None)?;
-            let step = step.as_ref(py);
+            let step = self.env.call_method_bound(py, "step", (action,), None)?;
+            let step = step.bind(py);
             let obs = step.get_item(0)?.call_method("flatten", (), None)?;
-            let obs_buffer = pyo3::buffer::PyBuffer::get(obs)?;
+            let obs_buffer = pyo3::buffer::PyBuffer::get_bound(&obs)?;
             let obs: Vec<u8> = obs_buffer.to_vec(py)?;
             let reward: Vec<f32> = step.get_item(1)?.extract()?;
             let is_done: Vec<f32> = step.get_item(2)?.extract()?;

candle-examples/examples/segformer/main.rs

@@ -5,7 +5,7 @@ use candle_transformers::models::segformer::{
     Config, ImageClassificationModel, SemanticSegmentationModel,
 };
 use clap::{Args, Parser, Subcommand};
-use image::Rgb;
+use imageproc::image::Rgb;
 use imageproc::integral_image::ArrayData;
 use std::collections::HashMap;
 use std::path::PathBuf;

candle-examples/examples/stable-diffusion/README.md

@@ -46,7 +46,8 @@ The default scheduler for the XL Turbo version is the Euler Ancestral scheduler.
 - `--cpu`: use the cpu rather than the gpu (much slower).
 - `--height`, `--width`: set the height and width for the generated image.
 - `--n-steps`: the number of steps to be used in the diffusion process.
-- `--num-samples`: the number of samples to generate.
+- `--num-samples`: the number of samples to generate iteratively.
+- `--bsize`: the numbers of samples to generate simultaneously.
 - `--final-image`: the filename for the generated image(s).
 
 ### Using flash-attention

candle-examples/examples/stable-diffusion/main.rs

@@ -9,6 +9,7 @@ use candle_transformers::models::stable_diffusion;
 use anyhow::{Error as E, Result};
 use candle::{DType, Device, IndexOp, Module, Tensor, D};
 use clap::Parser;
+use stable_diffusion::vae::AutoEncoderKL;
 use tokenizers::Tokenizer;
 
 #[derive(Parser)]
@@ -64,9 +65,13 @@ struct Args {
     #[arg(long)]
     n_steps: Option<usize>,
 
-    /// The number of samples to generate.
+    /// The number of samples to generate iteratively.
     #[arg(long, default_value_t = 1)]
-    num_samples: i64,
+    num_samples: usize,
+
+    /// The numbers of samples to generate simultaneously.
+    #[arg[long, default_value_t = 1]]
+    bsize: usize,
 
     /// The name of the final image to generate.
     #[arg(long, value_name = "FILE", default_value = "sd_final.png")]
@@ -236,8 +241,8 @@ impl ModelFile {
 
 fn output_filename(
     basename: &str,
-    sample_idx: i64,
-    num_samples: i64,
+    sample_idx: usize,
+    num_samples: usize,
     timestep_idx: Option<usize>,
 ) -> String {
     let filename = if num_samples > 1 {
@@ -261,6 +266,33 @@ fn output_filename(
     }
 }
 
+#[allow(clippy::too_many_arguments)]
+fn save_image(
+    vae: &AutoEncoderKL,
+    latents: &Tensor,
+    vae_scale: f64,
+    bsize: usize,
+    idx: usize,
+    final_image: &str,
+    num_samples: usize,
+    timestep_ids: Option<usize>,
+) -> Result<()> {
+    let images = vae.decode(&(latents / vae_scale)?)?;
+    let images = ((images / 2.)? + 0.5)?.to_device(&Device::Cpu)?;
+    let images = (images.clamp(0f32, 1.)? * 255.)?.to_dtype(DType::U8)?;
+    for batch in 0..bsize {
+        let image = images.i(batch)?;
+        let image_filename = output_filename(
+            final_image,
+            (bsize * idx) + batch + 1,
+            batch + num_samples,
+            timestep_ids,
+        );
+        candle_examples::save_image(&image, image_filename)?;
+    }
+    Ok(())
+}
+
 #[allow(clippy::too_many_arguments)]
 fn text_embeddings(
     prompt: &str,
@@ -292,6 +324,13 @@ fn text_embeddings(
         .map_err(E::msg)?
         .get_ids()
         .to_vec();
+    if tokens.len() > sd_config.clip.max_position_embeddings {
+        anyhow::bail!(
+            "the prompt is too long, {} > max-tokens ({})",
+            tokens.len(),
+            sd_config.clip.max_position_embeddings
+        )
+    }
     while tokens.len() < sd_config.clip.max_position_embeddings {
         tokens.push(pad_id)
     }
@@ -319,6 +358,13 @@ fn text_embeddings(
             .map_err(E::msg)?
             .get_ids()
             .to_vec();
+        if uncond_tokens.len() > sd_config.clip.max_position_embeddings {
+            anyhow::bail!(
+                "the negative prompt is too long, {} > max-tokens ({})",
+                uncond_tokens.len(),
+                sd_config.clip.max_position_embeddings
+            )
+        }
         while uncond_tokens.len() < sd_config.clip.max_position_embeddings {
             uncond_tokens.push(pad_id)
         }
@@ -368,6 +414,7 @@ fn run(args: Args) -> Result<()> {
         final_image,
         sliced_attention_size,
         num_samples,
+        bsize,
         sd_version,
         clip_weights,
         vae_weights,
@@ -461,6 +508,7 @@ fn run(args: Args) -> Result<()> {
         .collect::<Result<Vec<_>>>()?;
 
     let text_embeddings = Tensor::cat(&text_embeddings, D::Minus1)?;
+    let text_embeddings = text_embeddings.repeat((bsize, 1, 1))?;
     println!("{text_embeddings:?}");
 
     println!("Building the autoencoder.");
@@ -482,7 +530,6 @@ fn run(args: Args) -> Result<()> {
     } else {
         0
     };
-    let bsize = 1;
 
     let vae_scale = match sd_version {
         StableDiffusionVersion::V1_5
@@ -546,12 +593,16 @@ fn run(args: Args) -> Result<()> {
             println!("step {}/{n_steps} done, {:.2}s", timestep_index + 1, dt);
 
             if args.intermediary_images {
-                let image = vae.decode(&(&latents / vae_scale)?)?;
-                let image = ((image / 2.)? + 0.5)?.to_device(&Device::Cpu)?;
-                let image = (image * 255.)?.to_dtype(DType::U8)?.i(0)?;
-                let image_filename =
-                    output_filename(&final_image, idx + 1, num_samples, Some(timestep_index + 1));
-                candle_examples::save_image(&image, image_filename)?
+                save_image(
+                    &vae,
+                    &latents,
+                    vae_scale,
+                    bsize,
+                    idx,
+                    &final_image,
+                    num_samples,
+                    Some(timestep_index + 1),
+                )?;
             }
         }
 
@@ -560,11 +611,16 @@ fn run(args: Args) -> Result<()> {
             idx + 1,
             num_samples
         );
-        let image = vae.decode(&(&latents / vae_scale)?)?;
-        let image = ((image / 2.)? + 0.5)?.to_device(&Device::Cpu)?;
-        let image = (image.clamp(0f32, 1.)? * 255.)?.to_dtype(DType::U8)?.i(0)?;
-        let image_filename = output_filename(&final_image, idx + 1, num_samples, None);
-        candle_examples::save_image(&image, image_filename)?
+        save_image(
+            &vae,
+            &latents,
+            vae_scale,
+            bsize,
+            idx,
+            &final_image,
+            num_samples,
+            None,
+        )?;
     }
     Ok(())
 }

candle-examples/examples/stable-lm/main.rs

@@ -288,12 +288,12 @@ fn main() -> Result<()> {
     };
 
     let device = candle_examples::device(args.cpu)?;
-    let (model, device) = if args.quantized {
+    let model = if args.quantized {
         let filename = &filenames[0];
         let vb =
             candle_transformers::quantized_var_builder::VarBuilder::from_gguf(filename, &device)?;
         let model = QStableLM::new(&config, vb)?;
-        (Model::Quantized(model), Device::Cpu)
+        Model::Quantized(model)
     } else {
         let dtype = if device.is_cuda() {
             DType::BF16
@@ -302,7 +302,7 @@ fn main() -> Result<()> {
         };
         let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
         let model = StableLM::new(&config, vb)?;
-        (Model::StableLM(model), device)
+        Model::StableLM(model)
     };
 
     println!("loaded the model in {:?}", start.elapsed());

candle-examples/examples/t5/main.rs

@@ -12,12 +12,23 @@ use anyhow::{Error as E, Result};
 use candle::{DType, Device, Tensor};
 use candle_nn::VarBuilder;
 use candle_transformers::generation::LogitsProcessor;
-use clap::Parser;
+use clap::{Parser, ValueEnum};
 use hf_hub::{api::sync::Api, Repo, RepoType};
 use tokenizers::Tokenizer;
 
 const DTYPE: DType = DType::F32;
 
+#[derive(Clone, Debug, Copy, ValueEnum)]
+enum Which {
+    T5Base,
+    T5Small,
+    T5Large,
+    T5_3B,
+    Mt5Base,
+    Mt5Small,
+    Mt5Large,
+}
+
 #[derive(Parser, Debug, Clone)]
 #[command(author, version, about, long_about = None)]
 struct Args {
@@ -36,6 +47,15 @@ struct Args {
     #[arg(long)]
     revision: Option<String>,
 
+    #[arg(long)]
+    model_file: Option<String>,
+
+    #[arg(long)]
+    tokenizer_file: Option<String>,
+
+    #[arg(long)]
+    config_file: Option<String>,
+
     /// Enable decoding.
     #[arg(long)]
     decode: bool,
@@ -71,6 +91,10 @@ struct Args {
     /// The context size to consider for the repeat penalty.
     #[arg(long, default_value_t = 64)]
     repeat_last_n: usize,
+
+    /// The model to be used.
+    #[arg(long, default_value = "t5-small")]
+    which: Which,
 }
 
 struct T5ModelBuilder {
@@ -82,8 +106,17 @@ struct T5ModelBuilder {
 impl T5ModelBuilder {
     pub fn load(args: &Args) -> Result<(Self, Tokenizer)> {
         let device = candle_examples::device(args.cpu)?;
-        let default_model = "t5-small".to_string();
-        let default_revision = "refs/pr/15".to_string();
+        let (default_model, default_revision) = match args.which {
+            Which::T5Base => ("t5-base", "main"),
+            Which::T5Small => ("t5-small", "refs/pr/15"),
+            Which::T5Large => ("t5-large", "main"),
+            Which::T5_3B => ("t5-3b", "main"),
+            Which::Mt5Base => ("google/mt5-base", "refs/pr/5"),
+            Which::Mt5Small => ("google/mt5-small", "refs/pr/6"),
+            Which::Mt5Large => ("google/mt5-large", "refs/pr/2"),
+        };
+        let default_model = default_model.to_string();
+        let default_revision = default_revision.to_string();
         let (model_id, revision) = match (args.model_id.to_owned(), args.revision.to_owned()) {
             (Some(model_id), Some(revision)) => (model_id, revision),
             (Some(model_id), None) => (model_id, "main".to_string()),
@@ -93,14 +126,35 @@ impl T5ModelBuilder {
 
         let repo = Repo::with_revision(model_id.clone(), RepoType::Model, revision);
         let api = Api::new()?;
-        let api = api.repo(repo);
-        let config_filename = api.get("config.json")?;
-        let tokenizer_filename = api.get("tokenizer.json")?;
-        let weights_filename = if model_id == "google/flan-t5-xxl" || model_id == "google/flan-ul2"
-        {
-            candle_examples::hub_load_safetensors(&api, "model.safetensors.index.json")?
-        } else {
-            vec![api.get("model.safetensors")?]
+        let repo = api.repo(repo);
+        let config_filename = match &args.config_file {
+            None => repo.get("config.json")?,
+            Some(f) => f.into(),
+        };
+        let tokenizer_filename = match &args.tokenizer_file {
+            None => match args.which {
+                Which::Mt5Base => api
+                    .model("lmz/mt5-tokenizers".into())
+                    .get("mt5-base.tokenizer.json")?,
+                Which::Mt5Small => api
+                    .model("lmz/mt5-tokenizers".into())
+                    .get("mt5-small.tokenizer.json")?,
+                Which::Mt5Large => api
+                    .model("lmz/mt5-tokenizers".into())
+                    .get("mt5-large.tokenizer.json")?,
+                _ => repo.get("tokenizer.json")?,
+            },
+            Some(f) => f.into(),
+        };
+        let weights_filename = match &args.model_file {
+            Some(f) => f.split(',').map(|v| v.into()).collect::<Vec<_>>(),
+            None => {
+                if model_id == "google/flan-t5-xxl" || model_id == "google/flan-ul2" {
+                    candle_examples::hub_load_safetensors(&repo, "model.safetensors.index.json")?
+                } else {
+                    vec![repo.get("model.safetensors")?]
+                }
+            }
         };
         let config = std::fs::read_to_string(config_filename)?;
         let mut config: t5::Config = serde_json::from_str(&config)?;

candle-examples/examples/trocr/main.rs

@@ -115,7 +115,7 @@ pub fn main() -> anyhow::Result<()> {
     let processor = image_processor::ViTImageProcessor::new(&processor_config);
 
     let image = vec![args.image.as_str()];
-    let image = processor.preprocess(image)?;
+    let image = processor.preprocess(image)?.to_device(&device)?;
 
     let encoder_xs = model.encoder().forward(&image)?;
 

candle-examples/examples/whisper/README.md

@@ -34,6 +34,7 @@ from the hub.
 - `--timestamps`: enable the timestamp mode where some timestamps are reported
   for each recognized audio extracts.
 - `--model`: the model to be used. Models that do not end with `-en` are
-  multilingual models, other ones are English only models. The supported models
-  are `tiny`, `tiny.en`, `base`, `base.en`, `small`, `small.en`, `medium`,
-  `medium.en`, `large`, and `large-v2`.
+  multilingual models, other ones are English only models. The supported OpenAI 
+  Whisper models are `tiny`, `tiny.en`, `base`, `base.en`, `small`, `small.en`,
+  `medium`, `medium.en`, `large`, `large-v2` and `large-v3`. The supported 
+  Distil-Whisper models are `distil-medium.en`, `distil-large-v2` and `distil-large-v3`.

candle-examples/examples/whisper/main.rs

@@ -374,6 +374,8 @@ enum WhichModel {
     DistilMediumEn,
     #[value(name = "distil-large-v2")]
     DistilLargeV2,
+    #[value(name = "distil-large-v3")]
+    DistilLargeV3,
 }
 
 impl WhichModel {
@@ -386,7 +388,8 @@ impl WhichModel {
             | Self::Large
             | Self::LargeV2
             | Self::LargeV3
-            | Self::DistilLargeV2 => true,
+            | Self::DistilLargeV2
+            | Self::DistilLargeV3 => true,
             Self::TinyEn | Self::BaseEn | Self::SmallEn | Self::MediumEn | Self::DistilMediumEn => {
                 false
             }
@@ -408,6 +411,7 @@ impl WhichModel {
             Self::LargeV3 => ("openai/whisper-large-v3", "main"),
             Self::DistilMediumEn => ("distil-whisper/distil-medium.en", "main"),
             Self::DistilLargeV2 => ("distil-whisper/distil-large-v2", "main"),
+            Self::DistilLargeV3 => ("distil-whisper/distil-large-v3", "main"),
         }
     }
 }

candle-examples/examples/yolo-v8/main.rs

@@ -99,7 +99,7 @@ pub fn report_detect(
     let h_ratio = initial_h as f32 / h as f32;
     let mut img = img.to_rgb8();
     let font = Vec::from(include_bytes!("roboto-mono-stripped.ttf") as &[u8]);
-    let font = rusttype::Font::try_from_vec(font);
+    let font = ab_glyph::FontRef::try_from_slice(&font).map_err(candle::Error::wrap)?;
     for (class_index, bboxes_for_class) in bboxes.iter().enumerate() {
         for b in bboxes_for_class.iter() {
             println!(
@@ -119,27 +119,28 @@ pub fn report_detect(
                 );
             }
             if legend_size > 0 {
-                if let Some(font) = font.as_ref() {
-                    imageproc::drawing::draw_filled_rect_mut(
-                        &mut img,
-                        imageproc::rect::Rect::at(xmin, ymin).of_size(dx as u32, legend_size),
-                        image::Rgb([170, 0, 0]),
-                    );
-                    let legend = format!(
-                        "{}   {:.0}%",
-                        candle_examples::coco_classes::NAMES[class_index],
-                        100. * b.confidence
-                    );
-                    imageproc::drawing::draw_text_mut(
-                        &mut img,
-                        image::Rgb([255, 255, 255]),
-                        xmin,
-                        ymin,
-                        rusttype::Scale::uniform(legend_size as f32 - 1.),
-                        font,
-                        &legend,
-                    )
-                }
+                imageproc::drawing::draw_filled_rect_mut(
+                    &mut img,
+                    imageproc::rect::Rect::at(xmin, ymin).of_size(dx as u32, legend_size),
+                    image::Rgb([170, 0, 0]),
+                );
+                let legend = format!(
+                    "{}   {:.0}%",
+                    candle_examples::coco_classes::NAMES[class_index],
+                    100. * b.confidence
+                );
+                imageproc::drawing::draw_text_mut(
+                    &mut img,
+                    image::Rgb([255, 255, 255]),
+                    xmin,
+                    ymin,
+                    ab_glyph::PxScale {
+                        x: legend_size as f32 - 1.,
+                        y: legend_size as f32 - 1.,
+                    },
+                    &font,
+                    &legend,
+                )
             }
         }
     }

candle-flash-attn/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "candle-flash-attn"
-version = "0.4.2"
+version = "0.5.0"
 edition = "2021"
 
 description = "Flash attention layer for the candle ML framework."
@@ -11,7 +11,7 @@ license = "MIT OR Apache-2.0"
 readme = "README.md"
 
 [dependencies]
-candle = { path = "../candle-core", features = ["cuda"], package = "candle-core", version = "0.4.2" }
+candle = { path = "../candle-core", features = ["cuda"], package = "candle-core", version = "0.5.0" }
 half = { version = "2.3.1", features = ["num-traits"] }
 
 [build-dependencies]

candle-kernels/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "candle-kernels"
-version = "0.4.2"
+version = "0.5.0"
 edition = "2021"
 
 description = "CUDA kernels for Candle"

candle-kernels/build.rs

@@ -1,5 +1,8 @@
 fn main() {
     println!("cargo:rerun-if-changed=build.rs");
+    println!("cargo:rerun-if-changed=src/compatibility.cuh");
+    println!("cargo:rerun-if-changed=src/cuda_utils.cuh");
+    println!("cargo:rerun-if-changed=src/binary_op_macros.cuh");
 
     let builder = bindgen_cuda::Builder::default();
     println!("cargo:info={builder:?}");

candle-kernels/src/affine.cu

@@ -13,7 +13,7 @@ extern "C" __global__ void FN_NAME(  \
 ) {  \
     const size_t *dims = info; \
     const size_t *strides = info + num_dims; \
-    if (is_contiguous(num_dims, dims, strides)) { \
+    if (info == nullptr || is_contiguous(num_dims, dims, strides)) { \
         for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
             TYPENAME x = inp ? inp[i] : out[i]; \
             out[i] = x * mul + add; \

candle-kernels/src/binary_op_macros.cuh

@@ -12,8 +12,8 @@ extern "C" __global__ void FN_NAME( \
     const size_t *dims = dims_and_strides; \
     const size_t *lhs_strides = dims_and_strides + 1 * num_dims; \
     const size_t *rhs_strides = dims_and_strides + 2 * num_dims; \
-    bool lhs_cont = is_contiguous(num_dims, dims, lhs_strides); \
-    bool rhs_cont = is_contiguous(num_dims, dims, rhs_strides); \
+    bool lhs_cont = dims_and_strides == nullptr || is_contiguous(num_dims, dims, lhs_strides); \
+    bool rhs_cont = dims_and_strides == nullptr || is_contiguous(num_dims, dims, rhs_strides); \
     if (lhs_cont && rhs_cont) { \
         for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
             TYPENAME x = lhs[i]; \

candle-kernels/src/cast.cu

@@ -11,7 +11,7 @@ __device__ void cast_(
 ) {
     const size_t *dims = info;
     const size_t *strides = info + num_dims;
-    if (is_contiguous(num_dims, dims, strides)) {
+    if (info == nullptr || is_contiguous(num_dims, dims, strides)) {
         for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) {
             out[i] = inp[i];
         }
@@ -34,7 +34,7 @@ __device__ void cast_through(
 ) {
     const size_t *dims = info;
     const size_t *strides = info + num_dims;
-    if (is_contiguous(num_dims, dims, strides)) {
+    if (info == nullptr || is_contiguous(num_dims, dims, strides)) {
         for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) {
             out[i] = static_cast<T>(static_cast<I>(inp[i]));
         }
@@ -83,6 +83,18 @@ CAST_OP(double,   __nv_bfloat16, cast_f64_bf16)
 CAST_THROUGH_OP(__nv_bfloat16, uint8_t, float, cast_bf16_u8)
 CAST_THROUGH_OP(__nv_bfloat16, __half,   float, cast_bf16_f16)
 CAST_THROUGH_OP(__half,   __nv_bfloat16, float, cast_f16_bf16)
+#else
+#include <cuda.h>
+#if CUDA_VERSION >= 11000
+CAST_OP(__nv_bfloat16, float,    cast_bf16_f32)
+CAST_OP(float,    __nv_bfloat16, cast_f32_bf16)
+CAST_THROUGH_OP(__nv_bfloat16, uint8_t, float, cast_bf16_u8)
+CAST_THROUGH_OP(__nv_bfloat16, __half,  float, cast_bf16_f16)
+CAST_THROUGH_OP(__nv_bfloat16, double,  float, cast_bf16_f64)
+CAST_THROUGH_OP(__half,   __nv_bfloat16, float, cast_f16_bf16)
+CAST_THROUGH_OP(double,   __nv_bfloat16, float, cast_f64_bf16)
+CAST_THROUGH_OP(uint8_t,   __nv_bfloat16, float, cast_u8_bf16)
+#endif
 #endif
 
 #if __CUDA_ARCH__ >= 530

candle-kernels/src/cuda_utils.cuh

@@ -14,7 +14,7 @@ __device__ bool is_contiguous(
     size_t acc = 1;
     for (unsigned int d = 0; d < num_dims; d++) {
         unsigned int dim_idx = num_dims - 1 - d;
-        if (acc != strides[dim_idx]) {
+        if (dims[dim_idx] > 1 && acc != strides[dim_idx]) {
             return false;
         }
         acc *= dims[dim_idx];

candle-kernels/src/indexing.cu

@@ -168,8 +168,10 @@ IS_OP(__half, uint8_t, is_u8_f16)
 GATHER_OP(__half, int64_t, gather_i64_f16)
 GATHER_OP(__half, uint32_t, gather_u32_f16)
 GATHER_OP(__half, uint8_t, gather_u8_f16)
+IA_OP(__half, int64_t, ia_i64_f16)
 IA_OP(__half, uint32_t, ia_u32_f16)
 IA_OP(__half, uint8_t, ia_u8_f16)
+SA_OP(__half, int64_t, sa_i64_f16)
 SA_OP(__half, uint32_t, sa_u32_f16)
 SA_OP(__half, uint8_t, sa_u8_f16)
 #endif

candle-kernels/src/reduce.cu

@@ -2,6 +2,7 @@
 #include <cmath>
 #include <stdint.h>
 
+#define WARP_SIZE 32
 const int BLOCK_SIZE = 1024;
 
 // TODO: Maybe add some fast_sum_f16_f32 variant that not only accumulate in f32
@@ -49,6 +50,59 @@ fast_sum(const size_t src_numel, const size_t el_to_sum_per_block,
     dst[dst_id] = shr[0];
 }
 
+static __device__ __forceinline__ float warp_reduce_sum(float x) {
+#pragma unroll
+    for (int mask = 16; mask > 0; mask >>= 1) {
+        x += __shfl_xor_sync(0xffffffff, x, mask, 32);
+    }
+    return x;
+}
+
+// RmsNorm implementation adapted from ggml, accumulation is made using f32.
+// https://github.com/ggerganov/llama.cpp/blob/d59bd97065cd7ded6c4ecab54b1d5e0b1b11e318/ggml-cuda.cu#L523
+template <typename T>
+__device__ void rmsnorm(const T * x, T * dst, const T * alpha, const int ncols, const float eps) {
+    const int row = blockIdx.x*blockDim.y + threadIdx.y;
+    const int tid = threadIdx.x;
+    const int block_size = blockDim.x;
+
+    float tmp = 0.0f; // partial sum for thread in warp
+
+    for (int col = tid; col < ncols; col += block_size) {
+        const float xi = static_cast<float>(x[row*ncols + col]);
+        tmp += xi * xi;
+    }
+
+    // sum up partial sums
+    tmp = warp_reduce_sum(tmp);
+    if (block_size > WARP_SIZE) {
+        __shared__ float s_sum[32];
+        int warp_id = threadIdx.x / WARP_SIZE;
+        int lane_id = threadIdx.x % WARP_SIZE;
+        if (lane_id == 0) {
+            s_sum[warp_id] = tmp;
+        }
+        __syncthreads();
+        tmp = s_sum[lane_id];
+        tmp = warp_reduce_sum(tmp);
+    }
+
+    const float mean = tmp / ncols;
+    const float scale = rsqrtf(mean + eps);
+
+    if (alpha == nullptr) {
+      for (int col = tid; col < ncols; col += block_size) {
+          dst[row*ncols + col] = static_cast<T>(scale * static_cast<float>(x[row*ncols + col]));
+      }
+    }
+    else {
+      for (int col = tid; col < ncols; col += block_size) {
+          float a = static_cast<float>(alpha[col]);
+          dst[row*ncols + col] = static_cast<T>(scale * static_cast<float>(x[row*ncols + col]) * a);
+      }
+    }
+}
+
 // Softmax implementation adapted from ggml.
 // https://github.com/ggerganov/llama.cpp/blob/d59bd97065cd7ded6c4ecab54b1d5e0b1b11e318/ggml-cuda.cu#L4159
 template <typename T, typename ACC>
@@ -93,6 +147,65 @@ __device__ void softmax(const T * x, T * dst, const int ncols) {
     }
 }
 
+template <typename T>
+__device__ void ropei(const T * src, const T * cos, const T * sin, T * dst, const uint32_t bh, const uint32_t td) {
+    const int idx = blockIdx.x * blockDim.x + threadIdx.x;
+    if (2 * idx >= bh * td) return;
+
+    uint32_t rope_idx = idx % (td / 2);
+    T c = cos[rope_idx];
+    T s = sin[rope_idx];
+
+    dst[2 * idx] = src[2 * idx] * c - src[2 * idx + 1] * s;
+    dst[2 * idx + 1] = src[2 * idx] * s + src[2 * idx + 1] * c;
+}
+
+template <typename T>
+__device__ void rope(const T * src, const T * cos, const T * sin, T * dst, const uint32_t bh, const uint32_t td, const uint32_t d) {
+    const int idx = blockIdx.x * blockDim.x + threadIdx.x;
+    if (2 * idx >= bh * td) return;
+
+    uint32_t i_bh = idx / (td / 2);
+    uint32_t i_td = idx - (td / 2) * i_bh;
+    uint32_t i_t = i_td / (d / 2);
+    uint32_t i_d = i_td - (d / 2) * i_t;
+    uint32_t i1 = i_bh * td + i_t * d + i_d;
+    uint32_t i2 = i1 + d / 2;
+    uint32_t i_cs = i_t * (d / 2) + i_d;
+    T c = cos[i_cs];
+    T s = sin[i_cs];
+
+    dst[i1] = src[i1] * c - src[i2] * s;
+    dst[i2] = src[i1] * s + src[i2] * c;
+}
+
+template <typename T>
+__device__ void rope_thd(
+    const T * src,
+    const T * cos,
+    const T * sin,
+    T * dst,
+    const uint32_t b,
+    const uint32_t t,
+    const uint32_t h,
+    const uint32_t d
+) {
+    const int idx = blockIdx.x * blockDim.x + threadIdx.x;
+    if (2 * idx >= b * t * h * d) return;
+
+    uint32_t i_bth = idx / (d / 2);
+    uint32_t i_d = idx - (d / 2) * i_bth;
+    uint32_t i_t = (i_bth / h) % t;
+    uint32_t i1 = i_bth * d + i_d;
+    uint32_t i2 = i1 + d / 2;
+    uint32_t i_cs = i_t * (d / 2) + i_d;
+    T c = cos[i_cs];
+    T s = sin[i_cs];
+
+    dst[i1] = src[i1] * c - src[i2] * s;
+    dst[i2] = src[i1] * s + src[i2] * c;
+}
+
 template <typename T>
 __device__ void
 fast_max(const size_t src_numel, const size_t el_to_sum_per_block,
@@ -341,14 +454,57 @@ fast_argmax(const size_t src_numel, const size_t el_to_sum_per_block,
     softmax<TYPENAME, ACC_TYPENAME>(src, dst, n_cols);                         \
   }                                                                            \
 
+#define RMSNORM_OP(TYPENAME, FN_NAME) \
+  extern "C" __global__ void FN_NAME(                                          \
+      const TYPENAME *src, TYPENAME *dst, const TYPENAME *alpha,               \
+      const int n_cols, const float eps) {                                     \
+    rmsnorm<TYPENAME>(src, dst, alpha, n_cols, eps);                           \
+  }                                                                            \
+
+#define ROPE_OP(TYPENAME, FN_NAME, FN_NAME_I, FN_NAME_THD) \
+  extern "C" __global__ void FN_NAME_I( \
+      const TYPENAME *src, \
+      const TYPENAME *cos, \
+      const TYPENAME *sin, \
+      TYPENAME *dst, \
+      const uint32_t bh, \
+      const uint32_t td) { \
+    ropei<TYPENAME>(src, cos, sin, dst, bh, td); \
+  } \
+  extern "C" __global__ void FN_NAME( \
+      const TYPENAME *src, \
+      const TYPENAME *cos, \
+      const TYPENAME *sin, \
+      TYPENAME *dst, \
+      const uint32_t bh, \
+      const uint32_t td, \
+      const uint32_t d) { \
+    rope<TYPENAME>(src, cos, sin, dst, bh, td, d); \
+  } \
+  extern "C" __global__ void FN_NAME_THD( \
+      const TYPENAME *src, \
+      const TYPENAME *cos, \
+      const TYPENAME *sin, \
+      TYPENAME *dst, \
+      const uint32_t b, \
+      const uint32_t t, \
+      const uint32_t h, \
+      const uint32_t d) { \
+    rope_thd<TYPENAME>(src, cos, sin, dst, b, t, h, d); \
+  } \
+
 #if __CUDA_ARCH__ >= 800
 SOFTMAX_OP(__nv_bfloat16, float, softmax_bf16)
+RMSNORM_OP(__nv_bfloat16, rmsnorm_bf16)
+ROPE_OP(__nv_bfloat16, rope_bf16, rope_i_bf16, rope_thd_bf16)
 SUM_OP(__nv_bfloat16, sum_bf16)
 FAST_OP(__nv_bfloat16, fast_min_bf16, fast_max_bf16, fast_argmin_bf16, fast_argmax_bf16, fast_sum_bf16)
 #endif
 
 #if __CUDA_ARCH__ >= 530
 SOFTMAX_OP(__half, float, softmax_f16)
+RMSNORM_OP(__half, rmsnorm_f16)
+ROPE_OP(__half, rope_f16, rope_i_f16, rope_thd_f16)
 SUM_OP(__half, sum_f16)
 FAST_OP(__half, fast_min_f16, fast_max_f16, fast_argmin_f16, fast_argmax_f16, fast_sum_f16)
 #endif
@@ -358,6 +514,10 @@ SUM_OP(double, sum_f64)
 SUM_OP(uint32_t, sum_u32)
 SOFTMAX_OP(float, float, softmax_f32)
 SOFTMAX_OP(double, double, softmax_f64)
+RMSNORM_OP(float, rmsnorm_f32)
+RMSNORM_OP(double, rmsnorm_f64)
+ROPE_OP(float, rope_f32, rope_i_f32, rope_thd_f32)
+ROPE_OP(double, rope_f64, rope_i_f64, rope_thd_f64)
 
 FAST_OP(float, fast_min_f32, fast_max_f32, fast_argmin_f32, fast_argmax_f32, fast_sum_f32)
 FAST_OP(double, fast_min_f64, fast_max_f64, fast_argmin_f64, fast_argmax_f64, fast_sum_f64)

candle-kernels/src/unary.cu

@@ -13,7 +13,7 @@ extern "C" __global__ void FN_NAME( \
 ) { \
     const size_t *dims = info; \
     const size_t *strides = info + num_dims; \
-    if (is_contiguous(num_dims, dims, strides)) { \
+    if (info == nullptr || is_contiguous(num_dims, dims, strides)) { \
         for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
             TYPENAME x = inp ? inp[i] : out[i]; \
             out[i] = FUNC; \
@@ -71,7 +71,7 @@ extern "C" __global__ void FN_NAME( \
 ) { \
     const size_t *dims = info; \
     const size_t *strides = info + num_dims; \
-    if (is_contiguous(num_dims, dims, strides)) { \
+    if (info == nullptr || is_contiguous(num_dims, dims, strides)) { \
         for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
             TYPENAME x = inp ? inp[i] : out[i]; \
             out[i] = FUNC; \
@@ -86,6 +86,11 @@ extern "C" __global__ void FN_NAME( \
     } \
 } \
 
+template<typename T>
+__device__ T sign_(T t) {
+  return static_cast<T>(t > static_cast<T>(0)) - static_cast<T>(t < static_cast<T>(0));
+}
+
 
 #if __CUDA_ARCH__ >= 800
 UNARY_OP(__nv_bfloat16, ucopy_bf16, x)
@@ -110,6 +115,7 @@ UNARY_OP(__nv_bfloat16, urelu_bf16, relu_fwd(x))
 UNARY_OP1(__nv_bfloat16, uelu_bf16, elu_fwd(x, param))
 UNARY_OP(__nv_bfloat16, usilu_bf16, silu_fwd(x))
 UNARY_OP1(__nv_bfloat16, upowf_bf16, powg(x, param))
+UNARY_OP(__nv_bfloat16, usign_bf16, sign_(x))
 #endif
 
 #if __CUDA_ARCH__ >= 530
@@ -135,6 +141,7 @@ UNARY_OP(__half, urelu_f16, relu_fwd(x))
 UNARY_OP1(__half, uelu_f16, elu_fwd(x, param))
 UNARY_OP(__half, usilu_f16, silu_fwd(x))
 UNARY_OP1(__half, upowf_f16, powg(x, param))
+UNARY_OP(__half, usign_f16, sign_(x))
 #endif
 
 UNARY_OP(uint8_t, ucopy_u8, x)
@@ -184,3 +191,5 @@ UNARY_OP(float, usilu_f32, silu_fwd(x))
 UNARY_OP(double, usilu_f64, silu_fwd(x))
 UNARY_OP1(float, upowf_f32, powg(x, param))
 UNARY_OP1(double, upowf_f64, powg(x, param))
+UNARY_OP(float, usign_f32, sign_(x))
+UNARY_OP(double, usign_f64, sign_(x))

candle-metal-kernels/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "candle-metal-kernels"
-version = "0.4.2"
+version = "0.5.0"
 edition = "2021"
 
 description = "Metal kernels for Candle"

candle-metal-kernels/src/binary.metal

@@ -60,21 +60,24 @@ BINARY(FN, half, half, NAME##_f16, NAME##_f16_strided); \
 BINARY(FN, uint32_t, uint32_t, NAME##_u32, NAME##_u32_strided); \
 BINARY(FN, uint8_t, uint8_t, NAME##_u8, NAME##_u8_strided);
 
-#define INT64_BINARY_OP(NAME, FN) \
-BINARY(FN, int64_t, int64_t, NAME##_i64, NAME##_i64_strided);
-
-#define BFLOAT_BINARY_OP(FN, NAME) \
-BINARY(FN, bfloat, bfloat, NAME##_bf16, NAME##_bf16_strided);
-
 #define BINARY_OP_OUT(NAME, FN) \
 BINARY(FN, float, uint8_t, NAME##_f32, NAME##_f32_strided); \
 BINARY(FN, half, uint8_t, NAME##_f16, NAME##_f16_strided); \
 BINARY(FN, uint32_t, uint8_t, NAME##_u32, NAME##_u32_strided); \
 BINARY(FN, uint8_t, uint8_t, NAME##_u8, NAME##_u8_strided);
 
+#define INT64_BINARY_OP(NAME, FN) \
+BINARY(FN, int64_t, int64_t, NAME##_i64, NAME##_i64_strided);
+
 #define INT64_BINARY_OP_OUT(NAME, FN) \
 BINARY(FN, int64_t, uint8_t, NAME##_i64, NAME##_i64_strided);
 
+#define BFLOAT_BINARY_OP(FN, NAME) \
+BINARY(FN, bfloat, bfloat, NAME##_bf16, NAME##_bf16_strided);
+
+#define BFLOAT_BINARY_OP_OUT(NAME, FN) \
+BINARY(FN, bfloat, uint8_t, NAME##_bf16, NAME##_bf16_strided);
+
 BINARY_OP(x + y, add)
 BINARY_OP(x - y, sub)
 BINARY_OP(x * y, mul)
@@ -112,4 +115,11 @@ BFLOAT_BINARY_OP(x * y, mul)
 BFLOAT_BINARY_OP(x / y, div)
 BFLOAT_BINARY_OP(MIN(x, y), min)
 BFLOAT_BINARY_OP(MAX(x, y), max)
+
+BFLOAT_BINARY_OP_OUT(eq, x == y)
+BFLOAT_BINARY_OP_OUT(ne, x != y)
+BFLOAT_BINARY_OP_OUT(le, x <= y)
+BFLOAT_BINARY_OP_OUT(lt, x < y)
+BFLOAT_BINARY_OP_OUT(ge, x >= y)
+BFLOAT_BINARY_OP_OUT(gt, x > y)
 #endif

candle-metal-kernels/src/cast.metal

@@ -72,27 +72,60 @@ kernel void FN_NAME_STRIDED( \
     output[tid] = static_cast<RIGHT_TYPENAME>(static_cast<IR_TYPENAME>(input[get_strided_index(tid, num_dims, dims, strides)])); \
 } \
 
+// u32
 CAST(cast_u32_f32, cast_u32_f32_strided, uint32_t, float)
 CAST(cast_u32_u8, cast_u32_u8_strided, uint32_t, uint8_t)
-CAST(cast_u8_u32, cast_u8_u32_strided, uint8_t, uint32_t)
-CAST(cast_u8_f32, cast_u8_f32_strided, uint8_t, float)
-CAST(cast_f16_f32, cast_f16_f32_strided, half, float)
-CAST(cast_f32_f16, cast_f32_f16_strided, float, half)
-
+CAST(cast_u32_f16, cast_u32_f16_strided, uint32_t, half)
 #if __METAL_VERSION__ >= 220
-CAST(cast_u8_i64, cast_u8_i64_strided, uint8_t, int64_t)
 CAST(cast_u32_i64, cast_u32_i64_strided, uint32_t, int64_t)
-CAST(cast_i64_f32, cast_i64_f32_strided, int64_t, float)
+#endif
+#if defined(__HAVE_BFLOAT__)
+CAST(cast_u32_bf16, cast_u32_bf16_strided, uint32_t, bfloat)
 #endif
 
+// u8
+CAST(cast_u8_u32, cast_u8_u32_strided, uint8_t, uint32_t)
+CAST(cast_u8_f32, cast_u8_f32_strided, uint8_t, float)
+CAST(cast_u8_f16, cast_u8_f16_strided, uint8_t, half)
+#if __METAL_VERSION__ >= 220
+CAST(cast_u8_i64, cast_u8_i64_strided, uint8_t, int64_t)
+#endif
+#if defined(__HAVE_BFLOAT__)
+CAST(cast_u8_bf16, cast_u8_bf16_strided, uint8_t, bfloat)
+#endif
+
+// f16
+CAST(cast_f16_f32, cast_f16_f32_strided, half, float)
+CAST(cast_f16_u8, cast_f16_u8_strided, half, uint8_t)
+CAST(cast_f16_u32, cast_f16_u32_strided, half, uint32_t)
+CAST(cast_f16_i64, cast_f16_i64_strided, half, int64_t)
+#if defined(__HAVE_BFLOAT__)
+CAST_THROUGH(cast_f16_bf16, cast_f16_bf16_strided, half, bfloat, float)
+#endif
+
+// i64
+CAST(cast_i64_f32, cast_i64_f32_strided, int64_t, float)
+CAST(cast_i64_u8, cast_i64_u8_strided, int64_t, uint8_t)
+CAST(cast_i64_u32, cast_i64_u32_strided, int64_t, uint32_t)
+CAST(cast_i64_f16, cast_i64_f16_strided, int64_t, half)
+#if defined(__HAVE_BFLOAT__)
+CAST_THROUGH(cast_i64_bf16, cast_i64_bf16_strided, int64_t, bfloat, float)
+#endif
+
+// f32
+CAST(cast_f32_f16, cast_f32_f16_strided, float, half)
+CAST(cast_f32_u32, cast_f32_u32_strided, float, uint32_t)
+CAST(cast_f32_u8, cast_f32_u8_strided, float, uint8_t)
+CAST(cast_f32_i64, cast_f32_i64_strided, float, int64_t)
+#if defined(__HAVE_BFLOAT__)
+CAST(cast_f32_bf16, cast_f32_bf16_strided, float, bfloat)
+#endif
+
+// bf16
 #if defined(__HAVE_BFLOAT__)
 CAST(cast_bf16_u32, cast_bf16_u32_strided, bfloat, uint32_t)
+CAST(cast_bf16_i64, cast_bf16_i64_strided, bfloat, int64_t)
 CAST(cast_bf16_f32, cast_bf16_f32_strided, bfloat, float)
-CAST(cast_u8_bf16, cast_u8_bf16_strided, uint8_t, bfloat)
-CAST(cast_u32_bf16, cast_u32_bf16_strided, uint32_t, bfloat)
-CAST(cast_f32_bf16, cast_f32_bf16_strided, float, bfloat)
-
 CAST_THROUGH(cast_bf16_u8, cast_bf16_u8_strided, bfloat, uint8_t, float)
 CAST_THROUGH(cast_bf16_f16, cast_bf16_f16_strided, bfloat, half, float)
-CAST_THROUGH(cast_f16_bf16, cast_f16_bf16_strided, half, bfloat, float)
 #endif

candle-metal-kernels/src/conv.metal

@@ -1,3 +1,9 @@
+#include <metal_stdlib>
+
+using namespace metal;
+
+#define MAX(x, y) ((x) > (y) ? (x) : (y))
+
 template <typename T>
 METAL_FUNC void im2col(
     constant size_t &dst_numel,
@@ -200,14 +206,331 @@ kernel void FN_NAME(  \
   upsample_nearest2d<TYPENAME>(w_out, h_out, w_scale, h_scale, dims, strides, src, dst, tid); \
 } \
 
+template <typename T, typename A>
+METAL_FUNC void avg_pool2d(
+    constant size_t &w_k,
+    constant size_t &h_k,
+    constant size_t &w_stride,
+    constant size_t &h_stride,
+    constant size_t *src_dims,
+    constant size_t *src_strides,
+    device const T *src,
+    device T *dst,
+    uint tid [[ thread_position_in_grid ]]
+) {
+  const size_t c = src_dims[1];
+  const size_t w_in = src_dims[2];
+  const size_t h_in = src_dims[3];
+
+  const size_t w_out = (w_in - w_k) / w_stride + 1;
+  const size_t h_out = (h_in - h_k) / h_stride + 1;
+  if (tid >= src_dims[0] * c * w_out * h_out) {
+    return;
+  }
+
+  const size_t b_idx = tid / (w_out * h_out * c);
+  const size_t c_idx = (tid / (w_out * h_out)) % c;
+  const size_t dst_w = (tid / h_out) % w_out;
+  const size_t dst_h = tid % h_out;
+
+  const size_t src_idx0 = b_idx * src_strides[0];
+  A d = 0;
+  for (size_t w_offset = 0; w_offset < w_k; ++w_offset) {
+    size_t src_w = w_stride * dst_w + w_offset;
+    if (src_w >= w_in){
+      continue;
+    }
+    for (size_t h_offset = 0; h_offset < h_k; ++h_offset) {
+      size_t src_h = h_stride * dst_h + h_offset;
+      if (src_h >= h_in) {
+        continue;
+      }
+      const size_t src_idx = src_idx0 + c_idx * src_strides[1] + src_w * src_strides[2] + src_h * src_strides[3];
+      d += static_cast<A>(src[src_idx]);
+    }
+  }
+  dst[tid] = static_cast<T>(d / (w_k * h_k));
+}
+
+#define AVGPOOL2D_OP(TYPENAME, TYPEACC, FN_NAME) \
+kernel void FN_NAME( \
+    constant size_t &w_k, \
+    constant size_t &h_k, \
+    constant size_t &w_s, \
+    constant size_t &h_s, \
+    constant size_t *src_dims, \
+    constant size_t *src_s, \
+    device const TYPENAME *src, \
+    device TYPENAME *dst, \
+    uint tid [[ thread_position_in_grid ]] \
+) { \
+  avg_pool2d<TYPENAME, TYPEACC>(w_k, h_k, w_s, h_s, src_dims, src_s, src, dst, tid); \
+} \
+
+template <typename T>
+METAL_FUNC void max_pool2d(
+    constant size_t &w_k,
+    constant size_t &h_k,
+    constant size_t &w_stride,
+    constant size_t &h_stride,
+    constant size_t *src_dims,
+    constant size_t *src_strides,
+    device const T *src,
+    device T *dst,
+    uint tid [[ thread_position_in_grid ]]
+) {
+  const size_t c = src_dims[1];
+  const size_t w_in = src_dims[2];
+  const size_t h_in = src_dims[3];
+
+  const size_t w_out = (w_in - w_k) / w_stride + 1;
+  const size_t h_out = (h_in - h_k) / h_stride + 1;
+  if (tid >= src_dims[0] * c * w_out * h_out) {
+    return;
+  }
+
+  const size_t b_idx = tid / (w_out * h_out * c);
+  const size_t c_idx = (tid / (w_out * h_out)) % c;
+  const size_t dst_w = (tid / h_out) % w_out;
+  const size_t dst_h = tid % h_out;
+
+  const size_t src_idx0 = b_idx * src_strides[0];
+  T d = 0;
+  bool set = false;
+  for (size_t w_offset = 0; w_offset < w_k; ++w_offset) {
+    size_t src_w = w_stride * dst_w + w_offset;
+    if (src_w >= w_in){
+      continue;
+    }
+    for (size_t h_offset = 0; h_offset < h_k; ++h_offset) {
+      size_t src_h = h_stride * dst_h + h_offset;
+      if (src_h >= h_in) {
+        continue;
+      }
+      const size_t src_idx = src_idx0 + c_idx * src_strides[1] + src_w * src_strides[2] + src_h * src_strides[3];
+      if (set) {
+        d = MAX(d, src[src_idx]);
+      }
+      else {
+        d = src[src_idx];
+        set = true;
+      }
+    }
+  }
+  dst[tid] = d;
+}
+
+#define MAXPOOL2D_OP(TYPENAME, FN_NAME) \
+kernel void FN_NAME( \
+    constant size_t &w_k, \
+    constant size_t &h_k, \
+    constant size_t &w_s, \
+    constant size_t &h_s, \
+    constant size_t *src_dims, \
+    constant size_t *src_s, \
+    device const TYPENAME *src, \
+    device TYPENAME *dst, \
+    uint tid [[ thread_position_in_grid ]] \
+) { \
+  max_pool2d<TYPENAME>(w_k, h_k, w_s, h_s, src_dims, src_s, src, dst, tid); \
+} \
+
+
+// Naive implementation of conv_transpose1d.
+template <typename T, typename A>
+METAL_FUNC void conv_transpose1d(
+    constant size_t &l_out,
+    constant size_t &stride,
+    constant size_t &padding,
+    constant size_t &out_padding,
+    constant size_t &dilation,
+    constant size_t *src_dims,
+    constant size_t *src_strides,
+    constant size_t *k_dims,
+    constant size_t *k_strides,
+    device const T *src,
+    device const T *k,
+    device T *dst,
+    uint tid [[ thread_position_in_grid ]]
+) {
+  // src: (b_size, c_in, l_in)
+  // kernel: (c_in, c_out, l_k)
+  const size_t l_k = k_dims[2];
+  const size_t c_out = k_dims[1];
+  const size_t c_in = src_dims[1];
+  const size_t l_in = src_dims[2];
+  if (tid >= src_dims[0] * c_out * l_out) {
+    return;
+  }
+
+  const size_t b_idx = tid / (l_out * c_out);
+  const size_t dst_c_idx = (tid / l_out) % c_out;
+  const size_t out_x = tid % l_out;
+
+  const size_t src_idx0 = b_idx * src_strides[0];
+  A d = 0;
+  for (int k_x = 0; k_x < (int)l_k; ++k_x) {
+      // let out_x = inp_x * p.stride + k_x * p.dilation - p.padding;
+      int inp_x_stride = (int)(out_x + padding) - k_x * dilation;
+      if (inp_x_stride < 0 || inp_x_stride % stride) {
+          continue;
+      }
+      int inp_x = inp_x_stride / stride;
+      if (inp_x >= l_in) continue;
+      for (size_t src_c_idx = 0; src_c_idx < c_in; ++src_c_idx) {
+          const size_t src_idx = src_idx0 + src_c_idx * src_strides[1] + inp_x * src_strides[2];
+          const size_t k_idx = src_c_idx * k_strides[0] + dst_c_idx * k_strides[1] + k_x * k_strides[2];
+          d += static_cast<A>(src[src_idx]) * static_cast<A>(k[k_idx]);
+      }
+  }
+  dst[tid] = static_cast<T>(d);
+}
+
+#define CONVT1D_OP(TYPENAME, TYPEACC, FN_NAME) \
+kernel void FN_NAME(  \
+    constant size_t &l_out, \
+    constant size_t &stride, \
+    constant size_t &padding, \
+    constant size_t &out_padding, \
+    constant size_t &dilation, \
+    constant size_t *src_dims, \
+    constant size_t *src_strides, \
+    constant size_t *k_dims, \
+    constant size_t *k_strides, \
+    device const TYPENAME *src, \
+    device const TYPENAME *k, \
+    device TYPENAME *dst, \
+    uint tid [[ thread_position_in_grid ]] \
+) {  \
+  conv_transpose1d<TYPENAME, TYPEACC>(l_out, stride, padding, out_padding, dilation, src_dims, src_strides, k_dims, k_strides, src, k, dst, tid); \
+} \
+
+template <typename T, typename A>
+METAL_FUNC void conv_transpose2d(
+  constant size_t &w_out,
+  constant size_t &h_out,
+  constant size_t &stride,
+  constant size_t &padding,
+  constant size_t &out_padding,
+  constant size_t &dilation,
+  constant size_t *input_dims,
+  constant size_t *input_stride,
+  constant size_t *k_dims,
+  constant size_t *k_stride,
+  device const T *src,
+  device const T *k,
+  device T *dst,
+  uint tid [[ thread_position_in_grid ]]
+) {
+  const size_t h_k = k_dims[2];
+  const size_t w_k = k_dims[3];
+  const size_t c_out = k_dims[1];
+  const size_t c_in = input_dims[1];
+  const size_t h_in = input_dims[2];
+  const size_t w_in = input_dims[3];
+  
+  if (tid >= input_dims[0] * c_out * w_out * h_out) {
+    return;
+  }
+
+  const size_t b_idx = tid / (w_out * h_out * c_out);
+  const size_t dst_c_idx = (tid / (w_out * h_out)) % c_out;
+  const size_t out_y = (tid / w_out) % h_out;
+  const size_t out_x = tid % w_out;
+
+  const size_t src_idx0 = b_idx * input_stride[0];
+
+  A d = 0;
+  for (int k_x = 0; k_x < (int)w_k; ++k_x) {
+      const int inp_x_stride = (int)(out_x + padding) - k_x * dilation;
+      if (inp_x_stride < 0 || inp_x_stride % stride) {
+          continue;
+      }
+      const int inp_x = inp_x_stride / stride;
+      if (inp_x >= w_in) continue;
+      for (int k_y = 0; k_y < (int)h_k; ++k_y) {
+          const int inp_y_stride = (int)(out_y + padding) - k_y * dilation;
+          if (inp_y_stride < 0 || inp_y_stride % stride) {
+              continue;
+          }
+          const int inp_y = inp_y_stride / stride;
+          if (inp_y >= h_in) continue;
+          for (size_t src_c_idx = 0; src_c_idx < c_in; ++src_c_idx) {
+              const size_t src_idx = src_idx0 + src_c_idx * input_stride[1] + inp_y * input_stride[2] + inp_x * input_stride[3];
+              const size_t k_idx = src_c_idx * k_stride[0] + dst_c_idx * k_stride[1] + k_y * k_stride[2] + k_x * k_stride[3];
+              d += static_cast<A>(src[src_idx]) * static_cast<A>(k[k_idx]);
+          }
+      }
+  }
+  dst[tid] = static_cast<T>(d);
+}
+
+#define CONVT2D_OP(TYPENAME, TYPEACC, FN_NAME) \
+kernel void FN_NAME(  \
+    constant size_t &w_out, \
+    constant size_t &h_out, \
+    constant size_t &stride, \
+    constant size_t &padding, \
+    constant size_t &out_padding, \
+    constant size_t &dilation, \
+    constant size_t *input_dims, \
+    constant size_t *input_stride, \
+    constant size_t *k_dims, \
+    constant size_t *k_stride, \
+    device const TYPENAME *src, \
+    device const TYPENAME *k, \
+    device TYPENAME *dst, \
+    uint tid [[ thread_position_in_grid ]] \
+) {  \
+  conv_transpose2d<TYPENAME, TYPEACC>(w_out, h_out, stride, padding, out_padding, dilation, input_dims, input_stride, k_dims, k_stride, src, k, dst, tid); \
+} \
+
 IM2COL_OP(float, im2col_f32)
+IM2COL_OP(half, im2col_f16)
 IM2COL_OP(uint8_t, im2col_u8)
 IM2COL_OP(uint32_t, im2col_u32)
+#if defined(__HAVE_BFLOAT__)
+IM2COL_OP(bfloat, im2col_bf16)
+#endif
 
 IM2COL1D_OP(float, im2col1d_f32)
 IM2COL1D_OP(uint8_t, im2col1d_u8)
 IM2COL1D_OP(uint32_t, im2col1d_u32)
 
 UPSAMPLE_NEAREST2D_OP(float, upsample_nearest2d_f32)
+UPSAMPLE_NEAREST2D_OP(half, upsample_nearest2d_f16)
 UPSAMPLE_NEAREST2D_OP(uint8_t, upsample_nearest2d_u8)
 UPSAMPLE_NEAREST2D_OP(uint32_t, upsample_nearest2d_u32)
+#if defined(__HAVE_BFLOAT__)
+UPSAMPLE_NEAREST2D_OP(bfloat, upsample_nearest2d_bf16)
+#endif
+
+MAXPOOL2D_OP(float, max_pool2d_f32)
+MAXPOOL2D_OP(half, max_pool2d_f16)
+MAXPOOL2D_OP(uint32_t, max_pool2d_u32)
+MAXPOOL2D_OP(uint8_t, max_pool2d_u8)
+#if defined(__HAVE_BFLOAT__)
+MAXPOOL2D_OP(bfloat, max_pool2d_bf16)
+#endif
+
+AVGPOOL2D_OP(float, float, avg_pool2d_f32)
+AVGPOOL2D_OP(half, float, avg_pool2d_f16)
+AVGPOOL2D_OP(uint32_t, uint32_t, avg_pool2d_u32)
+AVGPOOL2D_OP(uint8_t, uint8_t, avg_pool2d_u8)
+#if defined(__HAVE_BFLOAT__)
+AVGPOOL2D_OP(bfloat, float, avg_pool2d_bf16)
+#endif
+
+CONVT1D_OP(float, float, conv_transpose1d_f32)
+CONVT1D_OP(half, float, conv_transpose1d_f16)
+CONVT1D_OP(uint8_t, uint8_t, conv_transpose1d_u8)
+CONVT1D_OP(uint32_t, uint32_t, conv_transpose1d_u32)
+#if defined(__HAVE_BFLOAT__)
+CONVT1D_OP(bfloat, float, conv_transpose1d_bf16)
+#endif
+
+CONVT2D_OP(float, float, conv_transpose2d_f32)
+CONVT2D_OP(half, float, conv_transpose2d_f16)
+#if defined(__HAVE_BFLOAT__)
+CONVT1D_OP(bfloat, float, conv_transpose2d_bf16)
+#endif

candle-metal-kernels/src/indexing.metal

@@ -1,20 +1,38 @@
 #include <metal_stdlib>
 using namespace metal;
 
+METAL_FUNC uint get_strided_index(
+    uint idx,
+    constant size_t &num_dims,
+    constant size_t *dims,
+    constant size_t *strides
+) {
+    uint strided_i = 0;
+    for (uint d = 0; d < num_dims; d++) {
+        uint dim_idx = num_dims - 1 - d;
+        strided_i += (idx % dims[dim_idx]) * strides[dim_idx];
+        idx /= dims[dim_idx];
+    }
+    return strided_i;
+}
+
 template<typename TYPENAME, typename INDEX_TYPENAME>
 METAL_FUNC void index( 
     constant size_t &dst_size, 
     constant size_t &left_size, 
     constant size_t &src_dim_size, 
     constant size_t &right_size, 
-    constant size_t &ids_size, 
-    const device TYPENAME *input, 
+    constant size_t &ids_size,
+    constant bool &contiguous,
+    constant size_t *src_dims,
+    constant size_t *src_strides,
+    const device TYPENAME *input,
     const device INDEX_TYPENAME *input_ids, 
     device TYPENAME *output, 
     uint tid [[ thread_position_in_grid ]] 
 ) { 
     if (tid >= dst_size) { 
-        return; 
+        return;
     } 
     const size_t id_i = (tid / right_size) % ids_size; 
     const INDEX_TYPENAME input_i = min(input_ids[id_i], (INDEX_TYPENAME)(src_dim_size - 1)); 
@@ -26,7 +44,8 @@ METAL_FUNC void index(
     // No need to check for zero we're only allowing unsized. 
     */ 
     const size_t src_i = left_rank_i * src_dim_size * right_size + input_i * right_size + right_rank_i; 
-    output[tid] = input[src_i]; 
+    const size_t strided_src_i = contiguous ? src_i : get_strided_index(src_i, src_dim_size, src_dims, src_strides);
+    output[tid] = input[strided_src_i];
 }
 
 # define INDEX_OP(NAME, INDEX_TYPENAME, TYPENAME) \
@@ -36,12 +55,15 @@ kernel void NAME( \
     constant size_t &src_dim_size, \
     constant size_t &right_size, \
     constant size_t &ids_size, \
+    constant bool &contiguous, \
+    constant size_t *src_dims, \
+    constant size_t *src_strides, \
     const device TYPENAME *input, \
     const device INDEX_TYPENAME *input_ids, \
     device TYPENAME *output, \
     uint tid [[ thread_position_in_grid ]] \
 ) { \
-    index<TYPENAME, INDEX_TYPENAME>(dst_size, left_size, src_dim_size, right_size, ids_size, input, input_ids, output, tid); \
+    index<TYPENAME, INDEX_TYPENAME>(dst_size, left_size, src_dim_size, right_size, ids_size, contiguous, src_dims, src_strides, input, input_ids, output, tid); \
 }
 
 
@@ -165,11 +187,29 @@ kernel void NAME( \
 }
 
 
-INDEX_OP(is_u32_f32, uint, float)
-INDEX_OP(is_u32_f16, uint, half)
+INDEX_OP(is_i64_f32, int64_t, float)
+INDEX_OP(is_i64_f16, int64_t, half)
+#if defined(__HAVE_BFLOAT__)
+INDEX_OP(is_i64_bf16, int64_t, bfloat)
+#endif
+
+INDEX_OP(is_u32_f32, uint32_t, float)
+INDEX_OP(is_u32_f16, uint32_t, half)
+#if defined(__HAVE_BFLOAT__)
+INDEX_OP(is_u32_bf16, uint32_t, bfloat)
+#endif
+
+INDEX_OP(is_u8_f32, uint8_t, float)
+INDEX_OP(is_u8_f16, uint8_t, half)
+#if defined(__HAVE_BFLOAT__)
+INDEX_OP(is_u8_bf16, uint8_t, bfloat)
+#endif
 
 GATHER_OP(gather_u32_f32, uint, float)
 GATHER_OP(gather_u32_f16, uint, half)
+#if defined(__HAVE_BFLOAT__)
+GATHER_OP(gather_u32_bf16, uint, bfloat)
+#endif
 
 SCATTER_ADD_OP(sa_u32_f32, uint32_t, float)
 SCATTER_ADD_OP(sa_u8_f32, uint8_t, float)
@@ -177,34 +217,38 @@ SCATTER_ADD_OP(sa_i64_f32, int64_t, float)
 SCATTER_ADD_OP(sa_u32_f16, uint32_t, half)
 SCATTER_ADD_OP(sa_u8_f16, uint8_t, half)
 SCATTER_ADD_OP(sa_i64_f16, int64_t, half)
-
 #if defined(__HAVE_BFLOAT__)
-INDEX_OP(is_u32_bf16, uint32_t, bfloat)
-INDEX_OP(is_u8_bf16, uint8_t, bfloat)
-
-INDEX_ADD_OP(ia_i64_bf16, int64_t, bfloat)
-INDEX_ADD_OP(ia_u32_bf16, uint32_t, bfloat)
-INDEX_ADD_OP(ia_u8_bf16, uint8_t, bfloat)
-
 SCATTER_ADD_OP(sa_u32_bf16, uint32_t, bfloat)
 SCATTER_ADD_OP(sa_u8_bf16, uint8_t, bfloat)
 SCATTER_ADD_OP(sa_i64_bf16, int64_t, bfloat)
 #endif
 
-INDEX_ADD_OP(ia_u32_f16, uint32_t, half)
-INDEX_ADD_OP(ia_u8_f16, uint8_t, half)
-
+// i64
+INDEX_ADD_OP(ia_i64_f16, int64_t, half)
 INDEX_ADD_OP(ia_i64_f32, int64_t, float)
-INDEX_ADD_OP(ia_i64_u8, int64_t, uint8_t)
 INDEX_ADD_OP(ia_i64_i64, int64_t, int64_t)
 INDEX_ADD_OP(ia_i64_u32, int64_t, uint32_t)
+INDEX_ADD_OP(ia_i64_u8, int64_t, uint8_t)
+#if defined(__HAVE_BFLOAT__)
+INDEX_ADD_OP(ia_i64_bf16, int64_t, bfloat)
+#endif
 
+// u32
+INDEX_ADD_OP(ia_u32_f16, uint32_t, half)
 INDEX_ADD_OP(ia_u32_f32, uint32_t, float)
-INDEX_ADD_OP(ia_u32_u8, uint32_t, uint8_t)
 INDEX_ADD_OP(ia_u32_i64, uint32_t, int64_t)
 INDEX_ADD_OP(ia_u32_u32, uint32_t, uint32_t)
+INDEX_ADD_OP(ia_u32_u8, uint32_t, uint8_t)
+#if defined(__HAVE_BFLOAT__)
+INDEX_ADD_OP(ia_u32_bf16, uint32_t, bfloat)
+#endif
 
+// u8
+INDEX_ADD_OP(ia_u8_f16, uint8_t, half)
 INDEX_ADD_OP(ia_u8_f32, uint8_t, float)
-INDEX_ADD_OP(ia_u8_u8, uint8_t, uint8_t)
-INDEX_ADD_OP(ia_u8_u32, uint8_t, uint32_t)
 INDEX_ADD_OP(ia_u8_i64, uint8_t, int64_t)
+INDEX_ADD_OP(ia_u8_u32, uint8_t, uint32_t)
+INDEX_ADD_OP(ia_u8_u8, uint8_t, uint8_t)
+#if defined(__HAVE_BFLOAT__)
+INDEX_ADD_OP(ia_u8_bf16, uint8_t, bfloat)
+#endif

candle-metal-kernels/src/reduce.metal

@@ -21,6 +21,52 @@ METAL_FUNC uint get_strided_index(
 
 constant int THREADGROUP_SIZE = 2048;
 
+template<typename T>
+METAL_FUNC void argmin(
+    constant size_t &num_dims,
+    constant size_t *dims,
+    constant size_t *strides,
+    constant size_t &el_to_sum_per_block,
+    device const T *src,
+    device uint *dst,
+    uint id,
+    uint tid,
+    uint dst_id,
+    uint block_dim,
+    threadgroup T *shared_memory,
+    threadgroup uint *shared_indices
+) {
+    bool notset = true;
+    // Elements summed in this block range from dst_id * el_to_sum_per_block
+    // to (dst_id + 1) * el_to_sum_per_block.
+    size_t start_idx = dst_id * el_to_sum_per_block;
+    size_t stop_idx = start_idx + el_to_sum_per_block;
+    size_t idx = start_idx + tid;
+    while (idx < stop_idx) {
+        // TODO: Fast version for the contiguous case.
+        size_t strided_i = get_strided_index(idx, num_dims, dims, strides);
+        if (notset || src[strided_i] < shared_memory[tid]) {
+            shared_memory[tid] = src[strided_i];
+            /* Assume that the reduction takes place over the last dimension which is contiguous. */
+            shared_indices[tid] = idx % dims[num_dims - 1];
+            notset = false;
+        }
+        idx += block_dim;
+    }
+
+    threadgroup_barrier(mem_flags::mem_none);
+    // reduction in shared memory
+    for (uint s = block_dim / 2; s > 0; s >>= 1) {
+        if (tid < s && shared_memory[tid + s] < shared_memory[tid]) {
+            shared_indices[tid] = shared_indices[tid + s];
+            shared_memory[tid] = shared_memory[tid + s];
+        }  \
+        threadgroup_barrier(mem_flags::mem_none);
+    }
+    if (tid == 0) {
+    dst[dst_id] = shared_indices[0];
+    }
+}
 
 #define ARGMIN(NAME, T, MAXVALUE) \
 kernel void NAME( \
@@ -35,53 +81,63 @@ kernel void NAME( \
     uint dst_id [[ threadgroup_position_in_grid ]],  \
     uint block_dim [[ threads_per_threadgroup ]]  \
 ) {  \
-      \
-   threadgroup T shared_memory[THREADGROUP_SIZE];  \
-   threadgroup uint shared_indices[THREADGROUP_SIZE];  \
-       \
-   shared_memory[tid] = MAXVALUE;  \
-   shared_indices[tid] = 0xFFFFFFFF; \
-   bool notset = true; \
-   /*  \
-   // Elements summed in this block range from dst_id * el_to_sum_per_block   \
-   // to (dst_id + 1) * el_to_sum_per_block.  \
-   */  \
-   size_t start_idx = dst_id * el_to_sum_per_block;  \
-   size_t stop_idx = start_idx + el_to_sum_per_block;  \
-   size_t idx = start_idx + tid;  \
-   while (idx < stop_idx) {  \
-     /*  \
-     // TODO: Fast version for the contiguous case.  \
-     */  \
-     size_t strided_i = get_strided_index(idx, num_dims, dims, strides);  \
-     if (notset || src[strided_i] < shared_memory[tid]) {  \
-         shared_memory[tid] = src[strided_i];  \
-          /* Assume that the reduction takes place over the last dimension which is contiguous. */ \
-          shared_indices[tid] = idx % dims[num_dims - 1]; \
-          notset = false; \
-     }  \
-     idx += block_dim;  \
-   }  \
-       \
-   threadgroup_barrier(mem_flags::mem_none);  \
-     \
-   /*  \
-   // reduction in shared memory  \
-   */  \
-   for (uint s = block_dim / 2; s > 0; s >>= 1) {  \
-       if (tid < s && shared_memory[tid + s] < shared_memory[tid]) {  \
-           shared_indices[tid] = shared_indices[tid + s];  \
-           shared_memory[tid] = shared_memory[tid + s];  \
-       }  \
-       threadgroup_barrier(mem_flags::mem_none);  \
-   }  \
-     \
-     if (tid == 0){ \
-       dst[dst_id] = shared_indices[0];  \
-     } \
+    threadgroup T shared_memory[THREADGROUP_SIZE]; \
+    threadgroup uint shared_indices[THREADGROUP_SIZE]; \
+    shared_memory[tid] = MAXVALUE; \
+    shared_indices[tid] = 0xFFFFFFFF; \
+    argmin<T>(num_dims, dims, strides, el_to_sum_per_block, src, dst, id, tid, dst_id, block_dim, shared_memory, shared_indices); \
 } \
 
 
+template<typename T>
+METAL_FUNC void argmax(
+    constant size_t & num_dims,
+    constant size_t * dims,
+    constant size_t * strides,
+    constant size_t & el_to_sum_per_block,
+    device const T * src,
+    device uint * dst,
+    uint id,
+    uint tid,
+    uint dst_id,
+    uint block_dim,
+    threadgroup T * shared_memory,
+    threadgroup uint * shared_indices
+  ) {
+    // Elements summed in this block range from dst_id * el_to_sum_per_block 
+    // to (dst_id + 1) * el_to_sum_per_block.
+    size_t start_idx = dst_id * el_to_sum_per_block;
+    size_t stop_idx = start_idx + el_to_sum_per_block;
+    size_t idx = start_idx + tid;
+    bool notset = true;
+    while (idx < stop_idx) {
+        // TODO: Fast version for the contiguous case.
+        size_t strided_i = get_strided_index(idx, num_dims, dims, strides);
+        if (notset || shared_memory[tid] < src[strided_i]) {
+            shared_memory[tid] = src[strided_i];
+            shared_indices[tid] = idx % dims[num_dims - 1];
+            notset = false;
+        }
+        idx += block_dim;
+    }
+
+    threadgroup_barrier(mem_flags::mem_none);
+
+    // reduction in shared memory
+    for (uint s = block_dim / 2; s > 0; s >>= 1) {
+        if (tid < s && shared_memory[tid + s] > shared_memory[tid]) {
+            shared_indices[tid] = shared_indices[tid + s];
+            shared_memory[tid] = shared_memory[tid + s];
+        }
+        threadgroup_barrier(mem_flags::mem_none);
+    }
+
+    // Thread 0 writes the result of the reduction
+    if (tid == 0) {
+        dst[dst_id] = shared_indices[0];
+    }
+  }
+
 #define ARGMAX(NAME, T, MINVALUE) \
 kernel void NAME( \
     constant size_t &num_dims, \
@@ -95,170 +151,337 @@ kernel void NAME( \
     uint dst_id [[ threadgroup_position_in_grid ]],  \
     uint block_dim [[ threads_per_threadgroup ]]  \
 ) {  \
-      \
    threadgroup T shared_memory[THREADGROUP_SIZE];  \
    threadgroup uint shared_indices[THREADGROUP_SIZE];  \
-       \
    shared_memory[tid] = MINVALUE;  \
    shared_indices[tid] = 0xFFFFFFFF; \
-   /*  \
-   // Elements summed in this block range from dst_id * el_to_sum_per_block   \
-   // to (dst_id + 1) * el_to_sum_per_block.  \
-   */  \
-   size_t start_idx = dst_id * el_to_sum_per_block;  \
-   size_t stop_idx = start_idx + el_to_sum_per_block;  \
-   size_t idx = start_idx + tid;  \
-   bool notset = true; \
-   while (idx < stop_idx) {  \
-     /*  \
-     // TODO: Fast version for the contiguous case.  \
-     */  \
-     size_t strided_i = get_strided_index(idx, num_dims, dims, strides);  \
-     if (notset || shared_memory[tid] < src[strided_i]) {  \
-         shared_memory[tid] = src[strided_i];  \
-         shared_indices[tid] = idx % dims[num_dims - 1]; \
-         notset = false; \
-     }  \
-     idx += block_dim;  \
-   }  \
-       \
-   threadgroup_barrier(mem_flags::mem_none);  \
-     \
-   /*  \
-   // reduction in shared memory  \
-   */  \
-   for (uint s = block_dim / 2; s > 0; s >>= 1) {  \
-       if (tid < s && shared_memory[tid + s] > shared_memory[tid]) {  \
-           shared_indices[tid] = shared_indices[tid + s];  \
-           shared_memory[tid] = shared_memory[tid + s];  \
-       }  \
-       threadgroup_barrier(mem_flags::mem_none);  \
-   }  \
-     \
-   if (tid == 0){ \
-       dst[dst_id] = shared_indices[0];  \
-   } \
+   argmax<T>(num_dims, dims, strides, el_to_sum_per_block, src, dst, id, tid, dst_id, block_dim, shared_memory, shared_indices);  \
 } \
 
+template<typename T>
+METAL_FUNC void reduce(
+    constant size_t & num_dims,
+    constant size_t * dims,
+    constant size_t * strides,
+    constant size_t & el_to_sum_per_block,
+    device const T * src,
+    device T * dst,
+    uint id,
+    uint tid,
+    uint dst_id,
+    uint block_dim,
+    threadgroup T * shared_memory,
+    T (*fn)(T, T)
+) {
+    // Elements summed in this block range from dst_id * el_to_sum_per_block 
+    // to (dst_id + 1) * el_to_sum_per_block.
+    size_t start_idx = dst_id * el_to_sum_per_block;
+    size_t stop_idx = start_idx + el_to_sum_per_block;
+    size_t idx = start_idx + tid;
+    while (idx < stop_idx) {
+        // TODO: Fast version for the contiguous case.
+        size_t strided_i = get_strided_index(idx, num_dims, dims, strides);
+        T x = shared_memory[tid];
+        T y = src[strided_i];
+        shared_memory[tid] = fn(x, y);
+        idx += block_dim;
+    }
+
+    threadgroup_barrier(mem_flags::mem_none);
+
+    // reduction in shared memory
+    for (uint s = block_dim / 2; s > 0; s >>= 1) {
+        if (tid < s) {
+            T x = shared_memory[tid];
+            T y = shared_memory[tid + s];
+            shared_memory[tid] = fn(x, y);
+        }
+        threadgroup_barrier(mem_flags::mem_none);
+    }
+
+    if (tid == 0) {
+        dst[dst_id] = shared_memory[0];
+    }
+}
+
 #define REDUCE(FN, NAME, T, START) \
+METAL_FUNC T NAME##_##op(T x, T y) { return FN; } \
 kernel void NAME( \
     constant size_t &num_dims, \
     constant size_t *dims, \
     constant size_t *strides, \
     constant size_t &el_to_sum_per_block, \
-    device const T *src,  \
+    device const T *src, \
     device T *dst, \
     uint id [[ thread_position_in_grid ]], \
     uint tid [[ thread_index_in_threadgroup ]], \
     uint dst_id [[ threadgroup_position_in_grid ]], \
     uint block_dim [[ threads_per_threadgroup ]] \
 ) { \
-     \
-   threadgroup T shared_memory[THREADGROUP_SIZE]; \
-      \
-   shared_memory[tid] = START; \
-   /* \
-   // Elements summed in this block range from dst_id * el_to_sum_per_block  \
-   // to (dst_id + 1) * el_to_sum_per_block. \
-   */ \
-   size_t start_idx = dst_id * el_to_sum_per_block; \
-   size_t stop_idx = start_idx + el_to_sum_per_block; \
-   size_t idx = start_idx + tid; \
-   while (idx < stop_idx) { \
-     /* \
-     // TODO: Fast version for the contiguous case. \
-     */ \
-     size_t strided_i = get_strided_index(idx, num_dims, dims, strides); \
-     T x = shared_memory[tid]; \
-     T y = src[strided_i]; \
-     shared_memory[tid] = FN; \
-     idx += block_dim; \
-   } \
-      \
-   threadgroup_barrier(mem_flags::mem_none); \
-    \
-   /* \
-   // reduction in shared memory \
-   */ \
-   for (uint s = block_dim / 2; s > 0; s >>= 1) { \
-       if (tid < s) { \
-           T x = shared_memory[tid]; \
-           T y = shared_memory[tid + s]; \
-           shared_memory[tid] = FN; \
-       } \
-       threadgroup_barrier(mem_flags::mem_none); \
-   } \
-    \
-   dst[dst_id] = shared_memory[0]; \
+    threadgroup T shared_memory[THREADGROUP_SIZE]; \
+    shared_memory[tid] = START; \
+    reduce<T>(num_dims, dims, strides, el_to_sum_per_block, src, dst, id, tid, dst_id, block_dim, shared_memory, NAME##_##op); \
 } \
 
+template<typename T>
+METAL_FUNC void softmax(
+    constant size_t & src_numel,
+    constant size_t & el_to_sum_per_block,
+    device const T * src,
+    device T * dst,
+    uint id,
+    uint tid,
+    uint dst_id,
+    uint block_dim,
+    threadgroup float * shared_memory
+) {
+    size_t start_idx = dst_id * el_to_sum_per_block;
+    size_t stop_idx = min(start_idx + el_to_sum_per_block, src_numel);
+    size_t idx = start_idx + tid;
 
-#define SOFTMAX(NAME, T)                                                          \
-kernel void NAME(                                                                 \
-    constant size_t &src_numel,                                                   \
-    constant size_t &el_to_sum_per_block,                                         \
-    device const T *src,                                                          \
-    device T *dst,                                                                \
-                                                                                  \
-    uint id [[ thread_position_in_grid ]],                                        \
-    uint tid [[ thread_index_in_threadgroup ]],                                   \
-    uint dst_id [[ threadgroup_position_in_grid ]],                               \
-    uint block_dim [[ threads_per_threadgroup ]]                                  \
-) {                                                                               \
-    threadgroup float shared_memory[THREADGROUP_SIZE];                                \
-    shared_memory[tid] = -INFINITY;                                            \
-    size_t start_idx = dst_id * el_to_sum_per_block;                              \
-    size_t stop_idx = min(start_idx + el_to_sum_per_block, src_numel);            \
-    size_t idx = start_idx + tid;                                                 \
-                                                                                  \
-                                                                                  \
-    float tmp = -INFINITY; \
-    while (idx < stop_idx) {                                                      \
-        tmp = MAX(tmp, float(src[idx]));                   \
-        idx += block_dim;                                                         \
-    }                                                                             \
-    shared_memory[tid] = tmp; \
-                                                                                  \
-    threadgroup_barrier(mem_flags::mem_threadgroup);                              \
-                                                                                  \
-    for (uint s = block_dim / 2; s > 0; s >>= 1) {                                \
-        if (tid < s) {                                                            \
-            shared_memory[tid] = MAX(shared_memory[tid], shared_memory[tid + s]); \
-        }                                                                         \
-        threadgroup_barrier(mem_flags::mem_threadgroup);                              \
-    }                                                                             \
-                                                                                  \
-    /* wait for shared_memory[0] to be filled */ \
-    threadgroup_barrier(mem_flags::mem_threadgroup);                              \
-                                                                                  \
-    float _max = shared_memory[0];                                                    \
-                                                                                  \
-    /* prevent tid=0 from overwriting _max before other threads have written it */ \
-    threadgroup_barrier(mem_flags::mem_threadgroup);                              \
-    shared_memory[tid] = 0;                                                       \
-                                                                                  \
-    idx = start_idx + tid;                                                        \
-    while (idx < stop_idx) {                                                      \
-        const float val = exp(float(src[idx]) - _max);                                    \
-        dst[idx] = T(val);                                                           \
-        shared_memory[tid] += val;                                                \
-        idx += block_dim;                                                         \
-    }                                                                             \
-    threadgroup_barrier(mem_flags::mem_threadgroup);                              \
-    for (uint s = block_dim / 2; s > 0; s >>= 1) {                                \
-        if (tid < s) {                                                            \
-            shared_memory[tid] += shared_memory[tid + s];                         \
-        }                                                                         \
-        threadgroup_barrier(mem_flags::mem_threadgroup);                              \
-    }                                                                             \
-                                                                                  \
-    const T inv_acc = T(1.0/shared_memory[0]);                                         \
-    idx = start_idx + tid;                                                        \
-    while (idx < stop_idx) {                                                      \
-        dst[idx] *= inv_acc;                                                      \
-        idx += block_dim;                                                         \
-    }                                                                             \
-}                                                                                 \
+    float tmp = -INFINITY;
+    while (idx < stop_idx) {
+        tmp = MAX(tmp, float(src[idx]));
+        idx += block_dim;
+    }
+    shared_memory[tid] = tmp;
+
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    for (uint s = block_dim / 2; s > 0; s >>= 1) {
+        if (tid < s) {
+            shared_memory[tid] = MAX(shared_memory[tid], shared_memory[tid + s]);\
+        }
+        threadgroup_barrier(mem_flags::mem_threadgroup);
+    }
+
+    /* wait for shared_memory[0] to be filled */
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    float _max = shared_memory[0];
+
+    /* prevent tid=0 from overwriting _max before other threads have written it */
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+    shared_memory[tid] = 0;
+
+    idx = start_idx + tid;
+    while (idx < stop_idx) {
+        const float val = exp(float(src[idx]) - _max);
+        dst[idx] = T(val);
+        shared_memory[tid] += val;
+        idx += block_dim;
+    }
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+    for (uint s = block_dim / 2; s > 0; s >>= 1) {
+        if (tid < s) {
+            shared_memory[tid] += shared_memory[tid + s];
+        }
+        threadgroup_barrier(mem_flags::mem_threadgroup);
+    }
+
+    const T inv_acc = T(1.0 / shared_memory[0]);
+    idx = start_idx + tid;
+    while (idx < stop_idx) {
+        dst[idx] *= inv_acc;
+        idx += block_dim;
+    }
+}
+
+#define SOFTMAX(NAME, T) \
+kernel void NAME( \
+    constant size_t &src_numel, \
+    constant size_t &el_to_sum_per_block, \
+    device const T *src, \
+    device T *dst, \
+    uint id [[ thread_position_in_grid ]], \
+    uint tid [[ thread_index_in_threadgroup ]], \
+    uint dst_id [[ threadgroup_position_in_grid ]], \
+    uint block_dim [[ threads_per_threadgroup ]] \
+) { \
+    threadgroup float shared_memory[THREADGROUP_SIZE]; \
+    shared_memory[tid] = -INFINITY; \
+    softmax<T>(src_numel, el_to_sum_per_block, src, dst, id, tid, dst_id, block_dim, shared_memory); \
+} \
+
+template<typename T>
+METAL_FUNC void rmsnorm(
+    constant size_t & src_numel,
+    constant size_t & el_to_sum_per_block,
+    device const T * src,
+    device T * dst,
+    device const T * alpha,
+    constant float & eps,
+    uint id,
+    uint tid,
+    uint dst_id,
+    uint block_dim,
+    threadgroup float * shared_memory
+) {
+    size_t start_idx = dst_id * el_to_sum_per_block;
+    size_t stop_idx = min(start_idx + el_to_sum_per_block, src_numel);
+    size_t idx = start_idx + tid;
+
+    float tmp = 0;
+    while (idx < stop_idx) {
+        tmp = tmp + float(src[idx]) * float(src[idx]);
+        idx += block_dim;
+    }
+    shared_memory[tid] = tmp;
+
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    for (uint s = block_dim / 2; s > 0; s >>= 1) {
+        if (tid < s) {
+            shared_memory[tid] = shared_memory[tid] + shared_memory[tid + s];
+        }
+        threadgroup_barrier(mem_flags::mem_threadgroup);
+    }
+
+    /* wait for shared_memory[0] to be filled */
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    float norm = sqrt(shared_memory[0] / float(el_to_sum_per_block) + eps);
+    float inv_norm = 1.0f / norm;
+    idx = start_idx + tid;
+    while (idx < stop_idx) {
+        float val = float(src[idx]) * inv_norm;
+        if (alpha != nullptr) {
+            val *= float(alpha[idx - start_idx]);
+        }
+        dst[idx] = T(val);
+        idx += block_dim;
+    }
+}
+
+#define RMSNORM(NAME, T) \
+kernel void NAME( \
+    constant size_t &src_numel, \
+    constant size_t &el_to_sum_per_block, \
+    device const T *src, \
+    device T *dst, \
+    device const T *alpha, \
+    constant float &eps, \
+    uint id [[ thread_position_in_grid ]], \
+    uint tid [[ thread_index_in_threadgroup ]], \
+    uint dst_id [[ threadgroup_position_in_grid ]], \
+    uint block_dim [[ threads_per_threadgroup ]] \
+) { \
+    threadgroup float shared_memory[THREADGROUP_SIZE]; \
+    shared_memory[tid] = 0; \
+    rmsnorm<T>(src_numel, el_to_sum_per_block, src, dst, alpha, eps, id, tid, dst_id, block_dim, shared_memory); \
+} \
+
+template<typename T>
+METAL_FUNC void ropei(
+    constant size_t &bh,
+    constant size_t &td,
+    device const T *src,
+    device const T *cos,
+    device const T *sin,
+    device T *dst,
+    uint tid
+) {
+    if (2 * tid >= bh * td) {
+        return;
+    }
+    size_t rope_idx = tid % (td / 2);
+    T c = cos[rope_idx];
+    T s = sin[rope_idx];
+    dst[2 * tid] = src[2 * tid] * c - src[2 * tid + 1] * s;
+    dst[2 * tid + 1] = src[2 * tid] * s + src[2 * tid + 1] * c;
+}
+
+template<typename T>
+METAL_FUNC void rope(
+    constant size_t &bh,
+    constant size_t &td,
+    constant size_t &d,
+    device const T *src,
+    device const T *cos,
+    device const T *sin,
+    device T *dst,
+    uint idx
+) {
+    if (2 * idx >= bh * td) {
+        return;
+    }
+    size_t i_bh = idx / (td / 2);
+    size_t i_td = idx - (td / 2) * i_bh;
+    size_t i_t = i_td / (d / 2);
+    size_t i_d = i_td - (d / 2) * i_t;
+    size_t i1 = i_bh * td + i_t * d + i_d;
+    size_t i2 = i1 + d / 2;
+    size_t i_cs = i_t * (d / 2) + i_d;
+    T c = cos[i_cs];
+    T s = sin[i_cs];
+    dst[i1] = src[i1] * c - src[i2] * s;
+    dst[i2] = src[i1] * s + src[i2] * c;
+}
+
+template<typename T>
+METAL_FUNC void rope_thd(
+    constant size_t &b,
+    constant size_t &t,
+    constant size_t &h,
+    constant size_t &d,
+    device const T *src,
+    device const T *cos,
+    device const T *sin,
+    device T *dst,
+    uint idx
+) {
+    if (2 * idx >= b * t * h * d) {
+        return;
+    }
+    const size_t i_bth = idx / (d / 2);
+    const size_t i_d = idx - (d / 2) * i_bth;
+    const size_t i_t = (i_bth / h) % t;
+    const size_t i1 = i_bth * d + i_d;
+    const size_t i2 = i1 + d / 2;
+    const size_t i_cs = i_t * (d / 2) + i_d;
+     T c = cos[i_cs];
+    T s = sin[i_cs];
+    dst[i1] = src[i1] * c - src[i2] * s;
+    dst[i2] = src[i1] * s + src[i2] * c;
+}
+
+#define ROPE(FN_NAME, FN_NAME_I, FN_NAME_THD, TYPENAME) \
+kernel void FN_NAME_I( \
+    constant size_t &bh, \
+    constant size_t &td, \
+    device const TYPENAME *src,  \
+    device const TYPENAME *cos,  \
+    device const TYPENAME *sin,  \
+    device TYPENAME *dst, \
+    uint tid [[ thread_position_in_grid ]] \
+) { \
+    ropei<TYPENAME>(bh, td, src, cos, sin, dst, tid); \
+}\
+kernel void FN_NAME( \
+    constant size_t &bh, \
+    constant size_t &td, \
+    constant size_t &d, \
+    device const TYPENAME *src,  \
+    device const TYPENAME *cos,  \
+    device const TYPENAME *sin,  \
+    device TYPENAME *dst, \
+    uint idx [[ thread_position_in_grid ]] \
+) { \
+    rope<TYPENAME>(bh, td, d, src, cos, sin, dst, idx); \
+}\
+kernel void FN_NAME_THD( \
+    constant size_t &b, \
+    constant size_t &t, \
+    constant size_t &h, \
+    constant size_t &d, \
+    device const TYPENAME *src,  \
+    device const TYPENAME *cos,  \
+    device const TYPENAME *sin,  \
+    device TYPENAME *dst, \
+    uint idx [[ thread_position_in_grid ]] \
+) { \
+    rope_thd<TYPENAME>(b, t, h, d, src, cos, sin, dst, idx); \
+}\
 
 REDUCE(x + y, fast_sum_f32_strided, float, 0)
 REDUCE(x + y, fast_sum_u32_strided, uint, 0)
@@ -286,6 +509,10 @@ ARGMAX(fast_argmax_u8_strided, uint8_t, 0)
 
 SOFTMAX(softmax_f32, float)
 SOFTMAX(softmax_f16, half)
+RMSNORM(rmsnorm_f32, float)
+RMSNORM(rmsnorm_f16, half)
+ROPE(rope_f32, rope_i_f32, rope_thd_f32, float)
+ROPE(rope_f16, rope_i_f16, rope_thd_f16, half)
 
 #if __METAL_VERSION__ >= 220
 REDUCE(x + y, fast_sum_i64_strided, int64_t, 0)
@@ -297,10 +524,16 @@ ARGMAX(fast_argmax_i64_strided, int64_t, INT_MIN)
 
 #if defined(__HAVE_BFLOAT__)
 REDUCE(x + y, fast_sum_bf16, bfloat, 0)
+REDUCE(x + y, fast_sum_bf16_strided, half, 0)
 REDUCE(x * y, fast_mul_bf16, bfloat, 1)
+REDUCE(x * y, fast_mul_bf16_strided, bfloat, 1)
 REDUCE(MAX(x, y), fast_max_bf16, bfloat, -HUGE_VALBF)
+REDUCE(MAX(x, y), fast_max_bf16_strided, bfloat, -HUGE_VALBF)
 REDUCE(MIN(x, y), fast_min_bf16, bfloat, HUGE_VALBF)
+REDUCE(MIN(x, y), fast_min_bf16_strided, bfloat, HUGE_VALBF)
 ARGMIN(fast_argmin_bf16, bfloat, HUGE_VALBF)
 ARGMAX(fast_argmax_bf16, bfloat, -HUGE_VALBF)
 SOFTMAX(softmax_bf16, bfloat)
+RMSNORM(rmsnorm_bf16, bfloat)
+ROPE(rope_bf16, rope_i_bf16, rope_thd_bf16, bfloat)
 #endif

candle-metal-kernels/src/unary.metal

@@ -104,21 +104,17 @@ UNARY(NAME, bfloat, NAME##_bf16, NAME##_bf16_strided);
 
 #define COPY2D(FN_NAME, TYPENAME) \
 kernel void FN_NAME( \
-    constant size_t &d1, \
-    constant size_t &d2, \
-    constant size_t &src_s, \
-    constant size_t &dst_s, \
+    constant int64_t &d1, \
+    constant int64_t &d2, \
+    constant int64_t &src_s, \
+    constant int64_t &dst_s, \
     device const TYPENAME *input,  \
     device TYPENAME *output, \
-    uint tid [[ thread_position_in_grid ]] \
+    uint2 idx [[thread_position_in_grid]] \
 ) { \
-    if (tid >= d1 * d2) { \
-        return; \
-    } \
-    size_t idx1 = tid / d2; \
-    size_t idx2 = tid - idx1 * d2; \
-    size_t src_idx = idx1 * src_s + idx2; \
-    size_t dst_idx = idx1 * dst_s + idx2; \
+    if (idx.x >= d1 || idx.y >= d2) return; \
+    int64_t src_idx = idx.x * src_s + idx.y; \
+    int64_t dst_idx = idx.x * dst_s + idx.y; \
     output[dst_idx] = input[src_idx]; \
 }
 
@@ -145,6 +141,7 @@ UNARY_OP(erf)
 UNARY_OP(tanh)
 UNARY_OP(recip)
 UNARY_OP(relu)
+UNARY_OP(sign)
 UNARY(id, float, copy_f32, copy_f32_strided)
 UNARY(id, half, copy_f16, copy_f16_strided)
 UNARY(id, uint8_t, copy_u8, copy_u8_strided)
@@ -174,8 +171,9 @@ BFLOAT_UNARY_OP(erf)
 BFLOAT_UNARY_OP(tanh)
 BFLOAT_UNARY_OP(recip)
 BFLOAT_UNARY_OP(relu)
+BFLOAT_UNARY_OP(sign)
 
 UNARY(id, bfloat, copy_bf16, copy_bf16_strided)
 
-COPY2D(copy2d_bf64, bfloat)
+COPY2D(copy2d_bf16, bfloat)
 #endif

candle-metal-kernels/src/utils.rs

@@ -0,0 +1,162 @@
+use metal::{Buffer, ComputeCommandEncoderRef, ComputePipelineState, MTLSize};
+use std::ffi::c_void;
+
+/// Most kernels apply similarly across the tensors
+/// This creates a strategy that uses the maximum amount of threads per threadgroup (capped at the
+/// actual total buffer length).
+/// Then kernels can just do their op on their single point in the buffer.
+pub(crate) fn linear_split(pipeline: &ComputePipelineState, length: usize) -> (MTLSize, MTLSize) {
+    let size = length as u64;
+    let width = std::cmp::min(pipeline.max_total_threads_per_threadgroup(), size);
+    let count = (size + width - 1) / width;
+    let thread_group_count = MTLSize {
+        width: count,
+        height: 1,
+        depth: 1,
+    };
+
+    let thread_group_size = MTLSize {
+        width,
+        height: 1,
+        depth: 1,
+    };
+    (thread_group_count, thread_group_size)
+}
+
+// https://github.com/ml-explore/mlx/blob/bddf23f175726a57f0e443cd45518c0757daa166/mlx/backend/metal/utils.h#L96
+pub(crate) fn get_block_dims(dim0: u64, dim1: u64, dim2: u64) -> MTLSize {
+    let mut pows0 = 0u64;
+    let mut pows1 = 0u64;
+    let mut pows2 = 0u64;
+    let mut sum = 0u64;
+    loop {
+        let presum = sum;
+        // Check all the pows
+        if dim0 >= (1 << (pows0 + 1)) {
+            pows0 += 1;
+            sum += 1;
+        }
+        if sum == 10 {
+            break;
+        }
+        if dim1 >= (1 << (pows1 + 1)) {
+            pows1 += 1;
+            sum += 1;
+        }
+        if sum == 10 {
+            break;
+        }
+        if dim2 >= (1 << (pows2 + 1)) {
+            pows2 += 1;
+            sum += 1;
+        }
+        if sum == presum || sum == 10 {
+            break;
+        }
+    }
+    MTLSize {
+        width: 1 << pows0,
+        height: 1 << pows1,
+        depth: 1 << pows2,
+    }
+}
+
+pub(crate) fn set_param<P: EncoderParam>(
+    encoder: &ComputeCommandEncoderRef,
+    position: u64,
+    data: P,
+) {
+    <P as EncoderParam>::set_param(encoder, position, data)
+}
+
+/// Helper functions to create the various objects on the compute command encoder
+/// on a single line.
+/// Prevents getting wrong some arguments number and mixing length and size in bytes.
+pub(crate) trait EncoderParam {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self);
+}
+macro_rules! primitive {
+    ($type:ty) => {
+        impl EncoderParam for $type {
+            fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+                encoder.set_bytes(
+                    position,
+                    core::mem::size_of::<$type>() as u64,
+                    &data as *const $type as *const c_void,
+                );
+            }
+        }
+    };
+}
+primitive!(bool);
+primitive!(usize);
+primitive!(i32);
+primitive!(i64);
+primitive!(u32);
+primitive!(u64);
+primitive!(f32);
+
+pub struct BufferOffset<'a> {
+    pub buffer: &'a Buffer,
+    pub offset_in_bytes: usize,
+}
+
+impl<'a> BufferOffset<'a> {
+    pub fn zero_offset(buffer: &'a Buffer) -> Self {
+        Self {
+            buffer,
+            offset_in_bytes: 0,
+        }
+    }
+}
+
+impl<T> EncoderParam for &[T] {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+        encoder.set_bytes(
+            position,
+            core::mem::size_of_val(data) as u64,
+            data.as_ptr() as *const c_void,
+        );
+    }
+}
+
+impl EncoderParam for &Buffer {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+        encoder.set_buffer(position, Some(data), 0);
+    }
+}
+
+impl EncoderParam for (&Buffer, usize) {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+        encoder.set_buffer(position, Some(data.0), data.1 as u64);
+    }
+}
+
+impl<'a> EncoderParam for &BufferOffset<'a> {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+        encoder.set_buffer(position, Some(data.buffer), data.offset_in_bytes as u64);
+    }
+}
+
+impl EncoderParam for &mut Buffer {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+        encoder.set_buffer(position, Some(data), 0);
+    }
+}
+
+impl EncoderParam for (&mut Buffer, usize) {
+    fn set_param(encoder: &ComputeCommandEncoderRef, position: u64, data: Self) {
+        encoder.set_buffer(position, Some(data.0), data.1 as u64);
+    }
+}
+
+#[macro_export]
+macro_rules! set_params {
+    ($encoder:ident, ($($param:expr),+)) => (
+        let mut _index = 0;
+        $(
+            $crate::utils::set_param($encoder, _index, $param);
+            _index += 1;
+        )*
+    );
+}

candle-nn/Cargo.toml

@@ -25,6 +25,8 @@ candle-metal-kernels = { workspace = true, optional = true }
 [dev-dependencies]
 anyhow = { workspace = true }
 clap = { workspace = true }
+rand = { workspace = true }
+criterion = { workspace = true }
 
 [features]
 default = []
@@ -32,3 +34,7 @@ accelerate = ["dep:accelerate-src", "candle/accelerate"]
 cuda = ["candle/cuda"]
 mkl = ["dep:intel-mkl-src", "candle/mkl"]
 metal = ["candle/metal", "dep:candle-metal-kernels", "dep:metal"]
+
+[[bench]]
+name = "bench_main"
+harness = false

candle-nn/benches/bench_main.rs

@@ -0,0 +1,4 @@
+mod benchmarks;
+
+use criterion::criterion_main;
+criterion_main!(benchmarks::layer_norm::benches, benchmarks::conv::benches);

candle-nn/benches/benchmarks/conv.rs

@@ -0,0 +1,54 @@
+use crate::benchmarks::{BenchDevice, BenchDeviceHandler};
+use candle::{DType, Device, Module, Tensor};
+use candle_nn::{Conv2d, Conv2dConfig};
+use criterion::{black_box, criterion_group, Criterion};
+use std::time::Instant;
+
+const B: usize = 1;
+const C: usize = 1;
+const M: usize = 128;
+const K: usize = 128;
+const K_SIZE: usize = 3;
+
+fn run(input: Tensor, weight: Tensor, bias: Tensor, config: Conv2dConfig) {
+    Conv2d::new(weight, Some(bias), config)
+        .forward(&input)
+        .unwrap();
+}
+
+fn run_conv2d_benchmark(c: &mut Criterion, device: &Device, dtype: DType, name: &str) {
+    let weight = Tensor::ones((1, 1, K_SIZE, K_SIZE), dtype, device)
+        .unwrap()
+        .to_dtype(dtype)
+        .unwrap();
+    let bias = Tensor::zeros(K, dtype, device).unwrap();
+    let input = Tensor::ones((B, C, M, K), dtype, device).unwrap();
+
+    let mut group = c.benchmark_group(device.bench_name(name));
+    group.bench_function("iter", move |b| {
+        b.iter_custom(|iters| {
+            let start = Instant::now();
+            for _i in 0..iters {
+                run(
+                    black_box(input.clone()),
+                    black_box(weight.clone()),
+                    black_box(bias.clone()),
+                    Default::default(),
+                );
+            }
+            device.sync().unwrap();
+            start.elapsed()
+        })
+    });
+    group.finish();
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let device = BenchDeviceHandler::new().unwrap();
+    for d in device.devices {
+        run_conv2d_benchmark(c, &d, DType::F32, "conv2d_f32");
+        run_conv2d_benchmark(c, &d, DType::F16, "conv2d_f16");
+    }
+}
+
+criterion_group!(benches, criterion_benchmark);

candle-nn/benches/benchmarks/layer_norm.rs

@@ -0,0 +1,48 @@
+use crate::benchmarks::{BenchDevice, BenchDeviceHandler};
+use candle::{DType, Device, Module, Tensor};
+use candle_nn::LayerNorm;
+use criterion::{black_box, criterion_group, Criterion};
+use std::time::Instant;
+
+fn run(input: &Tensor, weight: &Tensor, bias: &Tensor) {
+    let _ = LayerNorm::new(weight.clone(), bias.clone(), 1e-5).forward(&input);
+}
+
+const B: usize = 1;
+const M: usize = 1024;
+const K: usize = 1024;
+
+fn run_layer_norm_benchmark(c: &mut Criterion, device: &Device, dtype: DType, name: &str) {
+    let elements = B * M * K;
+
+    let weight = Tensor::arange(0.0, elements as f32, device)
+        .unwrap()
+        .to_dtype(dtype)
+        .unwrap();
+    let bias = weight.ones_like().unwrap();
+    let input = weight.ones_like().unwrap();
+
+    let mut group = c.benchmark_group(device.bench_name(name));
+    group.bench_function("iter", move |b| {
+        b.iter_custom(|iters| {
+            let start = Instant::now();
+            for _i in 0..iters {
+                run(black_box(&input), black_box(&weight), black_box(&bias));
+            }
+            device.sync().unwrap();
+            start.elapsed()
+        })
+    });
+    group.finish();
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let device = BenchDeviceHandler::new().unwrap();
+    for d in device.devices {
+        run_layer_norm_benchmark(c, &d, DType::F32, "layer_norm_f32");
+        run_layer_norm_benchmark(c, &d, DType::BF16, "layer_norm_bf16");
+        run_layer_norm_benchmark(c, &d, DType::F16, "layer_norm_f16");
+    }
+}
+
+criterion_group!(benches, criterion_benchmark);

candle-nn/benches/benchmarks/mod.rs

@@ -0,0 +1,64 @@
+pub(crate) mod conv;
+pub(crate) mod layer_norm;
+
+use candle::{Device, Result};
+
+pub(crate) trait BenchDevice {
+    fn sync(&self) -> Result<()>;
+
+    fn bench_name<S: Into<String>>(&self, name: S) -> String;
+}
+
+impl BenchDevice for Device {
+    fn sync(&self) -> Result<()> {
+        match self {
+            Device::Cpu => Ok(()),
+            Device::Cuda(device) => {
+                #[cfg(feature = "cuda")]
+                return Ok(device.synchronize()?);
+                #[cfg(not(feature = "cuda"))]
+                panic!("Cuda device without cuda feature enabled: {:?}", device)
+            }
+            Device::Metal(device) => {
+                #[cfg(feature = "metal")]
+                return Ok(device.wait_until_completed()?);
+                #[cfg(not(feature = "metal"))]
+                panic!("Metal device without metal feature enabled: {:?}", device)
+            }
+        }
+    }
+
+    fn bench_name<S: Into<String>>(&self, name: S) -> String {
+        match self {
+            Device::Cpu => {
+                let cpu_type = if cfg!(feature = "accelerate") {
+                    "accelerate"
+                } else if cfg!(feature = "mkl") {
+                    "mkl"
+                } else {
+                    "cpu"
+                };
+                format!("{}_{}", cpu_type, name.into())
+            }
+            Device::Cuda(_) => format!("cuda_{}", name.into()),
+            Device::Metal(_) => format!("metal_{}", name.into()),
+        }
+    }
+}
+
+struct BenchDeviceHandler {
+    devices: Vec<Device>,
+}
+
+impl BenchDeviceHandler {
+    pub fn new() -> Result<Self> {
+        let mut devices = Vec::new();
+        if cfg!(feature = "metal") {
+            devices.push(Device::new_metal(0)?);
+        } else if cfg!(feature = "cuda") {
+            devices.push(Device::new_cuda(0)?);
+        }
+        devices.push(Device::Cpu);
+        Ok(Self { devices })
+    }
+}

candle-nn/src/lib.rs

@@ -12,6 +12,7 @@ pub mod loss;
 pub mod ops;
 pub mod optim;
 pub mod rnn;
+pub mod rotary_emb;
 pub mod sequential;
 pub mod var_builder;
 pub mod var_map;

candle-nn/src/ops.rs

@@ -1,4 +1,4 @@
-use candle::{CpuStorage, Layout, Result, Shape, Tensor};
+use candle::{CpuStorage, DType, Layout, Result, Shape, Tensor};
 use rayon::prelude::*;
 
 /// Applies the softmax function to the input tensor, rescaling the element so that elements on
@@ -180,11 +180,10 @@ impl candle::CustomOp1 for SoftmaxLastDim {
                     block_dim: (1, 32, 1),
                     shared_mem_bytes: 0,
                 };
-                let src = &src.slice(layout.start_offset()..);
                 let func = dev.get_or_load_func(&kernel_name::<T>("softmax"), kernels::REDUCE)?;
                 // SAFETY: Set later by running the kernel.
                 let dst = unsafe { dev.alloc::<T>(el) }.w()?;
-                let params = (src, &dst, n_cols as i32);
+                let params = (&src, &dst, n_cols as i32);
                 // SAFETY: ffi.
                 unsafe { func.launch(cfg, params) }.w()?;
                 Ok(dst)
@@ -207,7 +206,7 @@ impl candle::CustomOp1 for SoftmaxLastDim {
         storage: &candle::MetalStorage,
         layout: &Layout,
     ) -> Result<(candle::MetalStorage, Shape)> {
-        use candle::{backend::BackendStorage, DType};
+        use candle::backend::BackendStorage;
         let device = storage.device();
         let command_buffer = device.command_buffer()?;
         let kernels = device.kernels();
@@ -237,7 +236,7 @@ impl candle::CustomOp1 for SoftmaxLastDim {
             layout.start_offset() * storage.dtype().size_in_bytes(),
             &output,
         )
-        .unwrap();
+        .map_err(candle::Error::wrap)?;
         let newstorage =
             candle::MetalStorage::new(output, device.clone(), elem_count, storage.dtype());
         Ok((newstorage, layout.shape().clone()))
@@ -248,6 +247,215 @@ pub fn softmax_last_dim(xs: &Tensor) -> Result<Tensor> {
     xs.apply_op1_no_bwd(&SoftmaxLastDim)
 }
 
+#[derive(Debug, Clone)]
+struct RmsNorm {
+    eps: f32,
+}
+
+impl candle::CustomOp2 for RmsNorm {
+    fn name(&self) -> &'static str {
+        "rms-norm"
+    }
+
+    fn cpu_fwd(
+        &self,
+        s1: &CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+    ) -> Result<(CpuStorage, Shape)> {
+        use candle::backend::BackendStorage;
+
+        let eps = self.eps;
+        fn inner<
+            T: candle::WithDType
+                + num_traits::Float
+                + num_traits::AsPrimitive<f32>
+                + num_traits::FromPrimitive,
+        >(
+            src: &[T],
+            layout: &Layout,
+            alpha: &[T],
+            alpha_layout: &Layout,
+            eps: f32,
+        ) -> Result<(CpuStorage, Shape)> {
+            let src = match layout.contiguous_offsets() {
+                None => candle::bail!("input has to be contiguous"),
+                Some((o1, o2)) => &src[o1..o2],
+            };
+            let alpha = match alpha_layout.contiguous_offsets() {
+                None => candle::bail!("alpha has to be contiguous"),
+                Some((o1, o2)) => &alpha[o1..o2],
+            };
+            let el_count = layout.shape().elem_count();
+            let dims = layout.shape().dims();
+            let dim_m1 = dims[dims.len() - 1];
+            let mut dst = vec![T::zero(); el_count];
+            src.par_chunks(dim_m1)
+                .zip(dst.par_chunks_mut(dim_m1))
+                .for_each(|(src, dst)| {
+                    let sum2 = src
+                        .iter()
+                        .map(|&v| {
+                            let v = v.as_();
+                            v * v
+                        })
+                        .sum::<f32>();
+                    let m = (sum2 / dim_m1 as f32 + eps).sqrt();
+                    let m = T::from_f32(m).unwrap_or_else(T::nan);
+                    for ((d, s), alpha) in dst.iter_mut().zip(src.iter()).zip(alpha) {
+                        *d = *s / m * *alpha
+                    }
+                });
+            let storage = candle::WithDType::to_cpu_storage_owned(dst);
+            Ok((storage, Shape::from_dims(dims)))
+        }
+
+        use CpuStorage as C;
+        match (s1, s2) {
+            (C::BF16(s1), C::BF16(s2)) => inner::<half::bf16>(s1, l1, s2, l2, eps),
+            (C::F16(s1), C::F16(s2)) => inner::<half::f16>(s1, l1, s2, l2, eps),
+            (C::F32(s1), C::F32(s2)) => inner::<f32>(s1, l1, s2, l2, eps),
+            _ => candle::bail!("unsupported dtype for rmsnorm {:?}", s1.dtype()),
+        }
+    }
+
+    #[cfg(feature = "cuda")]
+    fn cuda_fwd(
+        &self,
+        s1: &candle::CudaStorage,
+        l1: &Layout,
+        s2: &candle::CudaStorage,
+        l2: &Layout,
+    ) -> Result<(candle::CudaStorage, Shape)> {
+        use candle::cuda_backend::cudarc::driver::{
+            CudaSlice, DeviceRepr, LaunchAsync, LaunchConfig,
+        };
+        use candle::cuda_backend::{kernel_name, kernels, Map2, WrapErr};
+        use candle::{CudaDevice, WithDType};
+
+        struct S {
+            eps: f32,
+        }
+        impl Map2 for S {
+            fn f<T: DeviceRepr + WithDType>(
+                &self,
+                src: &CudaSlice<T>,
+                layout: &Layout,
+                alpha: &CudaSlice<T>,
+                alpha_layout: &Layout,
+                dev: &CudaDevice,
+            ) -> Result<CudaSlice<T>> {
+                let src = match layout.contiguous_offsets() {
+                    None => candle::bail!("input has to be contiguous"),
+                    Some((o1, o2)) => src.slice(o1..o2),
+                };
+                let alpha = match alpha_layout.contiguous_offsets() {
+                    None => candle::bail!("alpha has to be contiguous"),
+                    Some((o1, o2)) => alpha.slice(o1..o2),
+                };
+                let el = layout.shape().elem_count();
+                let dims = layout.shape().dims();
+                let dim_m1 = dims[dims.len() - 1];
+                let (n_rows, n_cols) = (el / dim_m1, dim_m1);
+
+                let cfg = LaunchConfig {
+                    grid_dim: (n_rows as u32, 1, 1),
+                    block_dim: (1024, 1, 1),
+                    shared_mem_bytes: 0,
+                };
+                let func = dev.get_or_load_func(&kernel_name::<T>("rmsnorm"), kernels::REDUCE)?;
+                // SAFETY: Set later by running the kernel.
+                let dst = unsafe { dev.alloc::<T>(el) }.w()?;
+                let params = (&src, &dst, &alpha, n_cols as i32, self.eps);
+                // SAFETY: ffi.
+                unsafe { func.launch(cfg, params) }.w()?;
+                Ok(dst)
+            }
+        }
+
+        use candle::backend::BackendStorage;
+        let dev = s1.device();
+        let slice = S { eps: self.eps }.map(&s1.slice, l1, &s2.slice, l2, dev)?;
+        let dst = candle::cuda_backend::CudaStorage {
+            slice,
+            device: dev.clone(),
+        };
+        Ok((dst, l1.shape().clone()))
+    }
+
+    #[cfg(feature = "metal")]
+    fn metal_fwd(
+        &self,
+        s1: &candle::MetalStorage,
+        l1: &Layout,
+        s2: &candle::MetalStorage,
+        l2: &Layout,
+    ) -> Result<(candle::MetalStorage, Shape)> {
+        use candle::backend::BackendStorage;
+        let device = s1.device();
+        let command_buffer = device.command_buffer()?;
+        let kernels = device.kernels();
+        let name = match (s1.dtype(), s2.dtype()) {
+            (DType::F32, DType::F32) => "rmsnorm_f32",
+            (DType::F16, DType::F16) => "rmsnorm_f16",
+            (DType::BF16, DType::BF16) => "rmsnorm_bf16",
+            (dt1, dt2) => candle::bail!("rmsnorm is not implemented for {dt1:?} {dt2:?}"),
+        };
+
+        if !(l1.is_contiguous() && l2.is_contiguous()) {
+            candle::bail!("Non contiguous rmsnorm is not implemented");
+        }
+
+        let last_dim = l1.dims()[l1.shape().rank() - 1];
+        let elem_count = l1.shape().elem_count();
+        let output = device.new_buffer(elem_count, s1.dtype(), "rmsnorm")?;
+        candle_metal_kernels::call_rms_norm(
+            device.metal_device(),
+            &command_buffer,
+            kernels,
+            name,
+            elem_count,
+            last_dim,
+            self.eps,
+            s1.buffer(),
+            l1.start_offset() * s1.dtype().size_in_bytes(),
+            s2.buffer(),
+            l2.start_offset() * s2.dtype().size_in_bytes(),
+            &output,
+        )
+        .map_err(candle::Error::wrap)?;
+        let newstorage = candle::MetalStorage::new(output, device.clone(), elem_count, s1.dtype());
+        Ok((newstorage, l1.shape().clone()))
+    }
+}
+
+pub fn rms_norm_slow(x: &Tensor, alpha: &Tensor, eps: f32) -> Result<Tensor> {
+    let x_dtype = x.dtype();
+    let internal_dtype = match x_dtype {
+        DType::F16 | DType::BF16 => DType::F32,
+        d => d,
+    };
+    let hidden_size = x.dim(candle::D::Minus1)?;
+    let x = x.to_dtype(internal_dtype)?;
+    let norm_x = (x.sqr()?.sum_keepdim(candle::D::Minus1)? / hidden_size as f64)?;
+    let x_normed = x.broadcast_div(&(norm_x + eps as f64)?.sqrt()?)?;
+    x_normed.to_dtype(x_dtype)?.broadcast_mul(alpha)
+}
+
+pub fn rms_norm(xs: &Tensor, alpha: &Tensor, eps: f32) -> Result<Tensor> {
+    let hidden_size_xs = xs.dim(candle::D::Minus1)?;
+    let hidden_size_alpha = alpha.dims1()?;
+    if hidden_size_xs != hidden_size_alpha {
+        candle::bail!(
+            "shape mismatch in rms-norm {:?} {:?}",
+            xs.shape(),
+            alpha.shape()
+        )
+    }
+    xs.apply_op2_no_bwd(alpha, &RmsNorm { eps })
+}
+
 // https://pytorch.org/docs/stable/generated/torch.nn.PixelShuffle.html
 pub fn pixel_shuffle(xs: &Tensor, upscale_factor: usize) -> Result<Tensor> {
     let (b_size, c, h, w) = xs.dims4()?;

candle-nn/src/rnn.rs

@@ -31,7 +31,7 @@ pub trait RNN {
         let (_b_size, seq_len, _features) = input.dims3()?;
         let mut output = Vec::with_capacity(seq_len);
         for seq_index in 0..seq_len {
-            let input = input.i((.., seq_index, ..))?;
+            let input = input.i((.., seq_index, ..))?.contiguous()?;
             let state = if seq_index == 0 {
                 self.step(&input, init_state)?
             } else {

candle-nn/src/rotary_emb.rs

@@ -0,0 +1,730 @@
+use candle::{CpuStorage, Layout, Result, Shape, Tensor, D};
+use rayon::prelude::*;
+
+/// Interleaved variant of rotary embeddings.
+/// The x0 and x1 value are interleaved on the n_embd (= head_dim) dimension.
+/// The resulting y0 and y1 are also interleaved with:
+///   y0 = x0*cos - x1*sin
+///   y1 = x0*sin + x1*cos
+#[derive(Debug, Clone)]
+struct RotaryEmbI;
+
+impl candle::CustomOp3 for RotaryEmbI {
+    fn name(&self) -> &'static str {
+        "rotary-emb-int"
+    }
+
+    fn cpu_fwd(
+        &self,
+        s1: &CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+        s3: &CpuStorage,
+        l3: &Layout,
+    ) -> Result<(CpuStorage, Shape)> {
+        fn inner<T: candle::WithDType + num_traits::Float>(
+            src: &[T],
+            l_src: &Layout,
+            cos: &[T],
+            l_cos: &Layout,
+            sin: &[T],
+            l_sin: &Layout,
+        ) -> Result<(CpuStorage, Shape)> {
+            let src = match l_src.contiguous_offsets() {
+                None => candle::bail!("input src has to be contiguous"),
+                Some((o1, o2)) => &src[o1..o2],
+            };
+            let cos = match l_cos.contiguous_offsets() {
+                None => candle::bail!("input cos has to be contiguous"),
+                Some((o1, o2)) => &cos[o1..o2],
+            };
+            let sin = match l_sin.contiguous_offsets() {
+                None => candle::bail!("input sin has to be contiguous"),
+                Some((o1, o2)) => &sin[o1..o2],
+            };
+            let (b, h, t, d) = l_src.shape().dims4()?;
+            let el_count = b * h * t * d;
+            let mut dst = vec![T::zero(); el_count];
+            src.par_chunks(t * d)
+                .zip(dst.par_chunks_mut(t * d))
+                .for_each(|(src, dst)| {
+                    for i_over_2 in 0..t * d / 2 {
+                        let i = 2 * i_over_2;
+                        dst[i] = src[i] * cos[i_over_2] - src[i + 1] * sin[i_over_2];
+                        dst[i + 1] = src[i] * sin[i_over_2] + src[i + 1] * cos[i_over_2];
+                    }
+                });
+            let storage = candle::WithDType::to_cpu_storage_owned(dst);
+            Ok((storage, (b, h, t, d).into()))
+        }
+
+        use candle::backend::BackendStorage;
+        use CpuStorage::{BF16, F16, F32, F64};
+        match (s1, s2, s3) {
+            (BF16(s1), BF16(s2), BF16(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F16(s1), F16(s2), F16(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F32(s1), F32(s2), F32(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F64(s1), F64(s2), F64(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            _ => candle::bail!(
+                "unsupported dtype for rope {:?} {:?} {:?}",
+                s1.dtype(),
+                s2.dtype(),
+                s3.dtype()
+            ),
+        }
+    }
+
+    #[cfg(feature = "cuda")]
+    fn cuda_fwd(
+        &self,
+        s1: &candle::CudaStorage,
+        l1: &Layout,
+        s2: &candle::CudaStorage,
+        l2: &Layout,
+        s3: &candle::CudaStorage,
+        l3: &Layout,
+    ) -> Result<(candle::CudaStorage, Shape)> {
+        use candle::cuda_backend::cudarc::driver::{
+            CudaSlice, DeviceRepr, LaunchAsync, LaunchConfig,
+        };
+        use candle::cuda_backend::{kernel_name, kernels, WrapErr};
+        use candle::{CudaDevice, WithDType};
+
+        fn inner<T: DeviceRepr + WithDType>(
+            src: &CudaSlice<T>,
+            l_src: &Layout,
+            cos: &CudaSlice<T>,
+            l_cos: &Layout,
+            sin: &CudaSlice<T>,
+            l_sin: &Layout,
+            dev: &CudaDevice,
+        ) -> Result<CudaSlice<T>> {
+            let src = match l_src.contiguous_offsets() {
+                None => candle::bail!("src input has to be contiguous"),
+                Some((o1, o2)) => src.slice(o1..o2),
+            };
+            let cos = match l_cos.contiguous_offsets() {
+                None => candle::bail!("cos input has to be contiguous"),
+                Some((o1, o2)) => cos.slice(o1..o2),
+            };
+            let sin = match l_sin.contiguous_offsets() {
+                None => candle::bail!("sin input has to be contiguous"),
+                Some((o1, o2)) => sin.slice(o1..o2),
+            };
+            let (b, h, t, d) = l_src.shape().dims4()?;
+            let el = b * h * t * d;
+            let cfg = LaunchConfig::for_num_elems((el / 2) as u32);
+            let func = dev.get_or_load_func(&kernel_name::<T>("rope_i"), kernels::REDUCE)?;
+            // SAFETY: Set later by running the kernel.
+            let dst = unsafe { dev.alloc::<T>(el) }.w()?;
+            let params = (&src, &cos, &sin, &dst, (b * h) as u32, (t * d) as u32);
+            // SAFETY: ffi.
+            unsafe { func.launch(cfg, params) }.w()?;
+            Ok(dst)
+        }
+
+        use candle::backend::BackendStorage;
+        use candle::cuda_backend::CudaStorageSlice::{BF16, F16, F32, F64};
+        let dev = s1.device();
+        let slice = match (&s1.slice, &s2.slice, &s3.slice) {
+            (BF16(s1), BF16(s2), BF16(s3)) => BF16(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F16(s1), F16(s2), F16(s3)) => F16(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F32(s1), F32(s2), F32(s3)) => F32(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F64(s1), F64(s2), F64(s3)) => F64(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            _ => candle::bail!(
+                "unsupported dtype for rope {:?} {:?} {:?}",
+                s1.dtype(),
+                s2.dtype(),
+                s3.dtype()
+            ),
+        };
+        let dst = candle::cuda_backend::CudaStorage {
+            slice,
+            device: dev.clone(),
+        };
+        Ok((dst, l1.shape().clone()))
+    }
+
+    #[cfg(feature = "metal")]
+    fn metal_fwd(
+        &self,
+        src: &candle::MetalStorage,
+        l_src: &Layout,
+        cos: &candle::MetalStorage,
+        l_cos: &Layout,
+        sin: &candle::MetalStorage,
+        l_sin: &Layout,
+    ) -> Result<(candle::MetalStorage, Shape)> {
+        use candle::backend::BackendStorage;
+        let device = src.device();
+        let command_buffer = device.command_buffer()?;
+        let kernels = device.kernels();
+        if cos.dtype() != src.dtype() || sin.dtype() != src.dtype() {
+            candle::bail!(
+                "dtype mismatch in rope-i {:?} {:?} {:?}",
+                src.dtype(),
+                cos.dtype(),
+                sin.dtype()
+            )
+        }
+        let name = match src.dtype() {
+            candle::DType::F32 => "rope_i_f32",
+            candle::DType::F16 => "rope_i_f16",
+            candle::DType::BF16 => "rope_i_bf16",
+            dtype => candle::bail!("rope-i is not implemented for {dtype:?}"),
+        };
+        let (b, h, t, d) = l_src.shape().dims4()?;
+        let el = b * h * t * d;
+        let output = device.new_buffer(el, src.dtype(), "rope-i")?;
+        candle_metal_kernels::call_rope_i(
+            device.metal_device(),
+            &command_buffer,
+            kernels,
+            name,
+            b * h,
+            t * d,
+            src.buffer(),
+            l_src.start_offset() * src.dtype().size_in_bytes(),
+            cos.buffer(),
+            l_cos.start_offset() * cos.dtype().size_in_bytes(),
+            sin.buffer(),
+            l_sin.start_offset() * sin.dtype().size_in_bytes(),
+            &output,
+        )
+        .map_err(candle::Error::wrap)?;
+        let out = candle::MetalStorage::new(output, device.clone(), el, src.dtype());
+        Ok((out, l_src.shape().clone()))
+    }
+}
+
+pub fn rope_i(xs: &Tensor, cos: &Tensor, sin: &Tensor) -> Result<Tensor> {
+    let (_b_sz, _n_head, seq_len, n_embd) = xs.dims4()?;
+    let (cos_seq_len, cos_n_embd) = cos.dims2()?;
+    let (sin_seq_len, sin_n_embd) = cos.dims2()?;
+    if cos_n_embd * 2 != n_embd
+        || sin_n_embd * 2 != n_embd
+        || seq_len > cos_seq_len
+        || seq_len > sin_seq_len
+    {
+        candle::bail!(
+            "inconsistent last dim size in rope {:?} {:?} {:?}",
+            xs.shape(),
+            cos.shape(),
+            sin.shape()
+        )
+    }
+    if !xs.is_contiguous() {
+        candle::bail!("xs has to be contiguous in rope")
+    }
+    if !cos.is_contiguous() {
+        candle::bail!("cos has to be contiguous in rope")
+    }
+    if !sin.is_contiguous() {
+        candle::bail!("sin has to be contiguous in rope")
+    }
+    xs.apply_op3_no_bwd(cos, sin, &RotaryEmbI)
+}
+
+pub fn rope_i_slow(x: &Tensor, cos: &Tensor, sin: &Tensor) -> Result<Tensor> {
+    let (b_sz, n_head, seq_len, n_embd) = x.dims4()?;
+    let cos = cos
+        .narrow(0, 0, seq_len)?
+        .reshape((seq_len, n_embd / 2, 1))?;
+    let sin = sin
+        .narrow(0, 0, seq_len)?
+        .reshape((seq_len, n_embd / 2, 1))?;
+    let cos = cos.broadcast_as((b_sz, 1, seq_len, n_embd / 2, 1))?;
+    let sin = sin.broadcast_as((b_sz, 1, seq_len, n_embd / 2, 1))?;
+    let x = x.reshape((b_sz, n_head, seq_len, n_embd / 2, 2))?;
+    let x0 = x.narrow(D::Minus1, 0, 1)?;
+    let x1 = x.narrow(D::Minus1, 1, 1)?;
+    let y0 = (x0.broadcast_mul(&cos)? - x1.broadcast_mul(&sin)?)?;
+    let y1 = (x0.broadcast_mul(&sin)? + x1.broadcast_mul(&cos)?)?;
+    let rope = Tensor::cat(&[y0, y1], D::Minus1)?;
+    let rope = rope.flatten_from(D::Minus2)?;
+    Ok(rope)
+}
+
+/// Contiguous variant of rope embeddings.
+#[derive(Debug, Clone)]
+struct RotaryEmb;
+
+impl candle::CustomOp3 for RotaryEmb {
+    fn name(&self) -> &'static str {
+        "rotary-emb"
+    }
+
+    fn cpu_fwd(
+        &self,
+        s1: &CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+        s3: &CpuStorage,
+        l3: &Layout,
+    ) -> Result<(CpuStorage, Shape)> {
+        fn inner<T: candle::WithDType + num_traits::Float>(
+            src: &[T],
+            l_src: &Layout,
+            cos: &[T],
+            l_cos: &Layout,
+            sin: &[T],
+            l_sin: &Layout,
+        ) -> Result<(CpuStorage, Shape)> {
+            let src = match l_src.contiguous_offsets() {
+                None => candle::bail!("input src has to be contiguous"),
+                Some((o1, o2)) => &src[o1..o2],
+            };
+            let cos = match l_cos.contiguous_offsets() {
+                None => candle::bail!("input cos has to be contiguous"),
+                Some((o1, o2)) => &cos[o1..o2],
+            };
+            let sin = match l_sin.contiguous_offsets() {
+                None => candle::bail!("input sin has to be contiguous"),
+                Some((o1, o2)) => &sin[o1..o2],
+            };
+            let (b, h, t, d) = l_src.shape().dims4()?;
+            let el_count = b * h * t * d;
+            let mut dst = vec![T::zero(); el_count];
+            src.par_chunks(t * d)
+                .zip(dst.par_chunks_mut(t * d))
+                .for_each(|(src, dst)| {
+                    for i_t in 0..t {
+                        for i_d in 0..d / 2 {
+                            let i1 = i_t * d + i_d;
+                            let i2 = i1 + d / 2;
+                            let i_cs = i_t * (d / 2) + i_d;
+                            dst[i1] = src[i1] * cos[i_cs] - src[i2] * sin[i_cs];
+                            dst[i2] = src[i1] * sin[i_cs] + src[i2] * cos[i_cs];
+                        }
+                    }
+                });
+            let storage = candle::WithDType::to_cpu_storage_owned(dst);
+            Ok((storage, (b, h, t, d).into()))
+        }
+
+        use candle::backend::BackendStorage;
+        use CpuStorage::{BF16, F16, F32, F64};
+        match (s1, s2, s3) {
+            (BF16(s1), BF16(s2), BF16(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F16(s1), F16(s2), F16(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F32(s1), F32(s2), F32(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F64(s1), F64(s2), F64(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            _ => candle::bail!(
+                "unsupported dtype for rope {:?} {:?} {:?}",
+                s1.dtype(),
+                s2.dtype(),
+                s3.dtype()
+            ),
+        }
+    }
+
+    #[cfg(feature = "cuda")]
+    fn cuda_fwd(
+        &self,
+        s1: &candle::CudaStorage,
+        l1: &Layout,
+        s2: &candle::CudaStorage,
+        l2: &Layout,
+        s3: &candle::CudaStorage,
+        l3: &Layout,
+    ) -> Result<(candle::CudaStorage, Shape)> {
+        use candle::cuda_backend::cudarc::driver::{
+            CudaSlice, DeviceRepr, LaunchAsync, LaunchConfig,
+        };
+        use candle::cuda_backend::{kernel_name, kernels, WrapErr};
+        use candle::{CudaDevice, WithDType};
+
+        fn inner<T: DeviceRepr + WithDType>(
+            src: &CudaSlice<T>,
+            l_src: &Layout,
+            cos: &CudaSlice<T>,
+            l_cos: &Layout,
+            sin: &CudaSlice<T>,
+            l_sin: &Layout,
+            dev: &CudaDevice,
+        ) -> Result<CudaSlice<T>> {
+            let src = match l_src.contiguous_offsets() {
+                None => candle::bail!("src input has to be contiguous"),
+                Some((o1, o2)) => src.slice(o1..o2),
+            };
+            let cos = match l_cos.contiguous_offsets() {
+                None => candle::bail!("cos input has to be contiguous"),
+                Some((o1, o2)) => cos.slice(o1..o2),
+            };
+            let sin = match l_sin.contiguous_offsets() {
+                None => candle::bail!("sin input has to be contiguous"),
+                Some((o1, o2)) => sin.slice(o1..o2),
+            };
+            let (b, h, t, d) = l_src.shape().dims4()?;
+            let el = b * h * t * d;
+            let cfg = LaunchConfig::for_num_elems((el / 2) as u32);
+            let func = dev.get_or_load_func(&kernel_name::<T>("rope"), kernels::REDUCE)?;
+            // SAFETY: Set later by running the kernel.
+            let dst = unsafe { dev.alloc::<T>(el) }.w()?;
+            let params = (
+                &src,
+                &cos,
+                &sin,
+                &dst,
+                (b * h) as u32,
+                (t * d) as u32,
+                d as u32,
+            );
+            // SAFETY: ffi.
+            unsafe { func.launch(cfg, params) }.w()?;
+            Ok(dst)
+        }
+
+        use candle::backend::BackendStorage;
+        use candle::cuda_backend::CudaStorageSlice::{BF16, F16, F32, F64};
+        let dev = s1.device();
+        let slice = match (&s1.slice, &s2.slice, &s3.slice) {
+            (BF16(s1), BF16(s2), BF16(s3)) => BF16(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F16(s1), F16(s2), F16(s3)) => F16(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F32(s1), F32(s2), F32(s3)) => F32(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F64(s1), F64(s2), F64(s3)) => F64(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            _ => candle::bail!(
+                "unsupported dtype for rope {:?} {:?} {:?}",
+                s1.dtype(),
+                s2.dtype(),
+                s3.dtype()
+            ),
+        };
+        let dst = candle::cuda_backend::CudaStorage {
+            slice,
+            device: dev.clone(),
+        };
+        Ok((dst, l1.shape().clone()))
+    }
+
+    #[cfg(feature = "metal")]
+    fn metal_fwd(
+        &self,
+        src: &candle::MetalStorage,
+        l_src: &Layout,
+        cos: &candle::MetalStorage,
+        l_cos: &Layout,
+        sin: &candle::MetalStorage,
+        l_sin: &Layout,
+    ) -> Result<(candle::MetalStorage, Shape)> {
+        use candle::backend::BackendStorage;
+        let device = src.device();
+        let command_buffer = device.command_buffer()?;
+        let kernels = device.kernels();
+        if cos.dtype() != src.dtype() || sin.dtype() != src.dtype() {
+            candle::bail!(
+                "dtype mismatch in rope {:?} {:?} {:?}",
+                src.dtype(),
+                cos.dtype(),
+                sin.dtype()
+            )
+        }
+        let name = match src.dtype() {
+            candle::DType::F32 => "rope_f32",
+            candle::DType::F16 => "rope_f16",
+            candle::DType::BF16 => "rope_bf16",
+            dtype => candle::bail!("rope is not implemented for {dtype:?}"),
+        };
+        let (b, h, t, d) = l_src.shape().dims4()?;
+        let el = b * h * t * d;
+        let output = device.new_buffer(el, src.dtype(), "rope-i")?;
+        candle_metal_kernels::call_rope(
+            device.metal_device(),
+            &command_buffer,
+            kernels,
+            name,
+            b * h,
+            t * d,
+            d,
+            src.buffer(),
+            l_src.start_offset() * src.dtype().size_in_bytes(),
+            cos.buffer(),
+            l_cos.start_offset() * cos.dtype().size_in_bytes(),
+            sin.buffer(),
+            l_sin.start_offset() * sin.dtype().size_in_bytes(),
+            &output,
+        )
+        .map_err(candle::Error::wrap)?;
+        let out = candle::MetalStorage::new(output, device.clone(), el, src.dtype());
+        Ok((out, l_src.shape().clone()))
+    }
+}
+
+pub fn rope(xs: &Tensor, cos: &Tensor, sin: &Tensor) -> Result<Tensor> {
+    let (_b_sz, _n_head, seq_len, n_embd) = xs.dims4()?;
+    let (cos_seq_len, cos_n_embd) = cos.dims2()?;
+    let (sin_seq_len, sin_n_embd) = sin.dims2()?;
+    if cos_n_embd * 2 != n_embd
+        || sin_n_embd * 2 != n_embd
+        || seq_len > cos_seq_len
+        || seq_len > sin_seq_len
+    {
+        candle::bail!(
+            "inconsistent last dim size in rope {:?} {:?} {:?}",
+            xs.shape(),
+            cos.shape(),
+            sin.shape()
+        )
+    }
+    if !xs.is_contiguous() {
+        candle::bail!("xs has to be contiguous in rope")
+    }
+    if !cos.is_contiguous() {
+        candle::bail!("cos has to be contiguous in rope")
+    }
+    if !sin.is_contiguous() {
+        candle::bail!("sin has to be contiguous in rope")
+    }
+    xs.apply_op3_no_bwd(cos, sin, &RotaryEmb)
+}
+
+fn rotate_half(xs: &Tensor) -> Result<Tensor> {
+    let last_dim = xs.dim(D::Minus1)?;
+    let xs1 = xs.narrow(D::Minus1, 0, last_dim / 2)?;
+    let xs2 = xs.narrow(D::Minus1, last_dim / 2, last_dim - last_dim / 2)?;
+    Tensor::cat(&[&xs2.neg()?, &xs1], D::Minus1)
+}
+
+pub fn rope_slow(x: &Tensor, cos: &Tensor, sin: &Tensor) -> Result<Tensor> {
+    let (_b_sz, _h, seq_len, _n_embd) = x.dims4()?;
+    let cos = Tensor::cat(&[cos, cos], D::Minus1)?;
+    let sin = Tensor::cat(&[sin, sin], D::Minus1)?;
+    let cos = cos.narrow(0, 0, seq_len)?;
+    let sin = sin.narrow(0, 0, seq_len)?;
+    let cos = cos.unsqueeze(0)?.unsqueeze(0)?;
+    let sin = sin.unsqueeze(0)?.unsqueeze(0)?;
+    x.broadcast_mul(&cos)? + rotate_half(x)?.broadcast_mul(&sin)?
+}
+
+/// T (seqlen)/H (num-heads)/D (head-dim) contiguous variant of rope embeddings.
+#[derive(Debug, Clone)]
+struct RotaryEmbThd;
+
+impl candle::CustomOp3 for RotaryEmbThd {
+    fn name(&self) -> &'static str {
+        "rotary-emb"
+    }
+
+    fn cpu_fwd(
+        &self,
+        s1: &CpuStorage,
+        l1: &Layout,
+        s2: &CpuStorage,
+        l2: &Layout,
+        s3: &CpuStorage,
+        l3: &Layout,
+    ) -> Result<(CpuStorage, Shape)> {
+        fn inner<T: candle::WithDType + num_traits::Float>(
+            src: &[T],
+            l_src: &Layout,
+            cos: &[T],
+            l_cos: &Layout,
+            sin: &[T],
+            l_sin: &Layout,
+        ) -> Result<(CpuStorage, Shape)> {
+            let src = match l_src.contiguous_offsets() {
+                None => candle::bail!("input src has to be contiguous"),
+                Some((o1, o2)) => &src[o1..o2],
+            };
+            let cos = match l_cos.contiguous_offsets() {
+                None => candle::bail!("input cos has to be contiguous"),
+                Some((o1, o2)) => &cos[o1..o2],
+            };
+            let sin = match l_sin.contiguous_offsets() {
+                None => candle::bail!("input sin has to be contiguous"),
+                Some((o1, o2)) => &sin[o1..o2],
+            };
+            let (b, t, h, d) = l_src.shape().dims4()?;
+            let el_count = b * h * t * d;
+            let mut dst = vec![T::zero(); el_count];
+            src.par_chunks(t * h * d)
+                .zip(dst.par_chunks_mut(t * h * d))
+                .for_each(|(src, dst)| {
+                    for i_t in 0..t {
+                        for i_d in 0..d / 2 {
+                            let i_cs = i_t * (d / 2) + i_d;
+                            for i_h in 0..h {
+                                let i1 = i_t * h * d + i_h * d + i_d;
+                                let i2 = i1 + d / 2;
+                                dst[i1] = src[i1] * cos[i_cs] - src[i2] * sin[i_cs];
+                                dst[i2] = src[i1] * sin[i_cs] + src[i2] * cos[i_cs];
+                            }
+                        }
+                    }
+                });
+            let storage = candle::WithDType::to_cpu_storage_owned(dst);
+            Ok((storage, (b, t, h, d).into()))
+        }
+
+        use candle::backend::BackendStorage;
+        use CpuStorage::{BF16, F16, F32, F64};
+        match (s1, s2, s3) {
+            (BF16(s1), BF16(s2), BF16(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F16(s1), F16(s2), F16(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F32(s1), F32(s2), F32(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            (F64(s1), F64(s2), F64(s3)) => inner(s1, l1, s2, l2, s3, l3),
+            _ => candle::bail!(
+                "unsupported dtype for rope {:?} {:?} {:?}",
+                s1.dtype(),
+                s2.dtype(),
+                s3.dtype()
+            ),
+        }
+    }
+
+    #[cfg(feature = "cuda")]
+    fn cuda_fwd(
+        &self,
+        s1: &candle::CudaStorage,
+        l1: &Layout,
+        s2: &candle::CudaStorage,
+        l2: &Layout,
+        s3: &candle::CudaStorage,
+        l3: &Layout,
+    ) -> Result<(candle::CudaStorage, Shape)> {
+        use candle::cuda_backend::cudarc::driver::{
+            CudaSlice, DeviceRepr, LaunchAsync, LaunchConfig,
+        };
+        use candle::cuda_backend::{kernel_name, kernels, WrapErr};
+        use candle::{CudaDevice, WithDType};
+
+        fn inner<T: DeviceRepr + WithDType>(
+            src: &CudaSlice<T>,
+            l_src: &Layout,
+            cos: &CudaSlice<T>,
+            l_cos: &Layout,
+            sin: &CudaSlice<T>,
+            l_sin: &Layout,
+            dev: &CudaDevice,
+        ) -> Result<CudaSlice<T>> {
+            let src = match l_src.contiguous_offsets() {
+                None => candle::bail!("src input has to be contiguous"),
+                Some((o1, o2)) => src.slice(o1..o2),
+            };
+            let cos = match l_cos.contiguous_offsets() {
+                None => candle::bail!("cos input has to be contiguous"),
+                Some((o1, o2)) => cos.slice(o1..o2),
+            };
+            let sin = match l_sin.contiguous_offsets() {
+                None => candle::bail!("sin input has to be contiguous"),
+                Some((o1, o2)) => sin.slice(o1..o2),
+            };
+            let (b, t, h, d) = l_src.shape().dims4()?;
+            let el = b * h * t * d;
+            let cfg = LaunchConfig::for_num_elems((el / 2) as u32);
+            let func = dev.get_or_load_func(&kernel_name::<T>("rope_thd"), kernels::REDUCE)?;
+            // SAFETY: Set later by running the kernel.
+            let dst = unsafe { dev.alloc::<T>(el) }.w()?;
+            let params = (
+                &src, &cos, &sin, &dst, b as u32, t as u32, h as u32, d as u32,
+            );
+            // SAFETY: ffi.
+            unsafe { func.launch(cfg, params) }.w()?;
+            Ok(dst)
+        }
+
+        use candle::backend::BackendStorage;
+        use candle::cuda_backend::CudaStorageSlice::{BF16, F16, F32, F64};
+        let dev = s1.device();
+        let slice = match (&s1.slice, &s2.slice, &s3.slice) {
+            (BF16(s1), BF16(s2), BF16(s3)) => BF16(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F16(s1), F16(s2), F16(s3)) => F16(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F32(s1), F32(s2), F32(s3)) => F32(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            (F64(s1), F64(s2), F64(s3)) => F64(inner(s1, l1, s2, l2, s3, l3, dev)?),
+            _ => candle::bail!(
+                "unsupported dtype for rope {:?} {:?} {:?}",
+                s1.dtype(),
+                s2.dtype(),
+                s3.dtype()
+            ),
+        };
+        let dst = candle::cuda_backend::CudaStorage {
+            slice,
+            device: dev.clone(),
+        };
+        Ok((dst, l1.shape().clone()))
+    }
+
+    #[cfg(feature = "metal")]
+    fn metal_fwd(
+        &self,
+        src: &candle::MetalStorage,
+        l_src: &Layout,
+        cos: &candle::MetalStorage,
+        l_cos: &Layout,
+        sin: &candle::MetalStorage,
+        l_sin: &Layout,
+    ) -> Result<(candle::MetalStorage, Shape)> {
+        use candle::backend::BackendStorage;
+        let device = src.device();
+        let command_buffer = device.command_buffer()?;
+        let kernels = device.kernels();
+        if cos.dtype() != src.dtype() || sin.dtype() != src.dtype() {
+            candle::bail!(
+                "dtype mismatch in rope {:?} {:?} {:?}",
+                src.dtype(),
+                cos.dtype(),
+                sin.dtype()
+            )
+        }
+        let name = match src.dtype() {
+            candle::DType::F32 => "rope_thd_f32",
+            candle::DType::F16 => "rope_thd_f16",
+            candle::DType::BF16 => "rope_thd_bf16",
+            dtype => candle::bail!("rope_thd is not implemented for {dtype:?}"),
+        };
+        let (b, t, h, d) = l_src.shape().dims4()?;
+        let el = b * h * t * d;
+        let output = device.new_buffer(el, src.dtype(), "rope-thd")?;
+        candle_metal_kernels::call_rope_thd(
+            device.metal_device(),
+            &command_buffer,
+            kernels,
+            name,
+            b,
+            t,
+            h,
+            d,
+            src.buffer(),
+            l_src.start_offset() * src.dtype().size_in_bytes(),
+            cos.buffer(),
+            l_cos.start_offset() * cos.dtype().size_in_bytes(),
+            sin.buffer(),
+            l_sin.start_offset() * sin.dtype().size_in_bytes(),
+            &output,
+        )
+        .map_err(candle::Error::wrap)?;
+        let out = candle::MetalStorage::new(output, device.clone(), el, src.dtype());
+        Ok((out, l_src.shape().clone()))
+    }
+}
+
+pub fn rope_thd(xs: &Tensor, cos: &Tensor, sin: &Tensor) -> Result<Tensor> {
+    let (_b_sz, seq_len, _n_head, n_embd) = xs.dims4()?;
+    let (cos_seq_len, cos_n_embd) = cos.dims2()?;
+    let (sin_seq_len, sin_n_embd) = sin.dims2()?;
+    if cos_n_embd * 2 != n_embd
+        || sin_n_embd * 2 != n_embd
+        || seq_len > cos_seq_len
+        || seq_len > sin_seq_len
+    {
+        candle::bail!(
+            "inconsistent last dim size in rope {:?} {:?} {:?}",
+            xs.shape(),
+            cos.shape(),
+            sin.shape()
+        )
+    }
+    if !xs.is_contiguous() {
+        candle::bail!("xs has to be contiguous in rope")
+    }
+    if !cos.is_contiguous() {
+        candle::bail!("cos has to be contiguous in rope")
+    }
+    if !sin.is_contiguous() {
+        candle::bail!("sin has to be contiguous in rope")
+    }
+    xs.apply_op3_no_bwd(cos, sin, &RotaryEmbThd)
+}

candle-nn/src/var_builder.rs

@@ -178,16 +178,27 @@ impl<'a, B: Backend> VarBuilderArgs<'a, B> {
         name: &str,
         hints: B::Hints,
     ) -> Result<Tensor> {
-        let path = self.path(name);
-        self.data
-            .backend
-            .get(s.into(), &path, hints, self.data.dtype, &self.data.device)
+        self.get_with_hints_dtype(s, name, hints, self.data.dtype)
     }
 
     /// Retrieve the tensor associated with the given name at the current path.
     pub fn get<S: Into<Shape>>(&self, s: S, name: &str) -> Result<Tensor> {
         self.get_with_hints(s, name, Default::default())
     }
+
+    /// Retrieve the tensor associated with the given name & dtype at the current path.
+    pub fn get_with_hints_dtype<S: Into<Shape>>(
+        &self,
+        s: S,
+        name: &str,
+        hints: B::Hints,
+        dtype: DType,
+    ) -> Result<Tensor> {
+        let path = self.path(name);
+        self.data
+            .backend
+            .get(s.into(), &path, hints, dtype, &self.data.device)
+    }
 }
 
 struct Zeros;
@@ -487,6 +498,53 @@ impl<'a> VarBuilder<'a> {
         let pth = candle::pickle::PthTensors::new(p, None)?;
         Ok(Self::from_backend(Box::new(pth), dtype, dev.clone()))
     }
+
+    /// Gets a VarBuilder that applies some renaming function on tensor it gets queried for before
+    /// passing the new names to the inner VarBuilder.
+    ///
+    /// ```rust
+    /// use candle::{Tensor, DType, Device};
+    ///
+    /// let a = Tensor::arange(0f32, 6f32, &Device::Cpu)?.reshape((2, 3))?;
+    /// let tensors: std::collections::HashMap<_, _> = [
+    ///     ("foo".to_string(), a),
+    /// ]
+    /// .into_iter()
+    /// .collect();
+    /// let vb = candle_nn::VarBuilder::from_tensors(tensors, DType::F32, &Device::Cpu);
+    /// assert!(vb.contains_tensor("foo"));
+    /// assert!(vb.get((2, 3), "foo").is_ok());
+    /// assert!(!vb.contains_tensor("bar"));
+    /// let vb = vb.rename_f(|f: &str| if f == "bar" { "foo".to_string() } else { f.to_string() });
+    /// assert!(vb.contains_tensor("bar"));
+    /// assert!(vb.contains_tensor("foo"));
+    /// assert!(vb.get((2, 3), "bar").is_ok());
+    /// assert!(vb.get((2, 3), "foo").is_ok());
+    /// assert!(!vb.contains_tensor("baz"));
+    /// # Ok::<(), candle::Error>(())
+    /// ```
+    pub fn rename_f<F: Fn(&str) -> String + Sync + Send + 'static>(self, f: F) -> Self {
+        let f: Box<dyn Fn(&str) -> String + Sync + Send + 'static> = Box::new(f);
+        self.rename(f)
+    }
+
+    pub fn rename<R: Renamer + Send + Sync + 'a>(self, renamer: R) -> Self {
+        let dtype = self.dtype();
+        let device = self.device().clone();
+        let path = self.path.clone();
+        let backend = Rename::new(self, renamer);
+        let backend: Box<dyn SimpleBackend + 'a> = Box::new(backend);
+        let data = TensorData {
+            backend,
+            dtype,
+            device,
+        };
+        Self {
+            data: Arc::new(data),
+            path,
+            _phantom: std::marker::PhantomData,
+        }
+    }
 }
 
 pub struct ShardedSafeTensors(candle::safetensors::MmapedSafetensors);
@@ -607,3 +665,49 @@ impl Backend for ShardedSafeTensors {
         self.0.get(name).is_ok()
     }
 }
+
+/// This traits specifies a way to rename the queried names into names that are stored in an inner
+/// VarBuilder.
+pub trait Renamer {
+    /// This is applied to the name obtained by a name call and the resulting name is passed to the
+    /// inner VarBuilder.
+    fn rename(&self, v: &str) -> std::borrow::Cow<'_, str>;
+}
+
+pub struct Rename<'a, R: Renamer> {
+    inner: VarBuilder<'a>,
+    renamer: R,
+}
+
+impl<'a, R: Renamer + Sync + Send> SimpleBackend for Rename<'a, R> {
+    fn get(
+        &self,
+        s: Shape,
+        name: &str,
+        h: crate::Init,
+        dtype: DType,
+        dev: &Device,
+    ) -> Result<Tensor> {
+        let name = self.renamer.rename(name);
+        self.inner
+            .get_with_hints_dtype(s, &name, h, dtype)?
+            .to_device(dev)
+    }
+
+    fn contains_tensor(&self, name: &str) -> bool {
+        let name = self.renamer.rename(name);
+        self.inner.contains_tensor(&name)
+    }
+}
+
+impl<'a, R: Renamer> Rename<'a, R> {
+    pub fn new(inner: VarBuilder<'a>, renamer: R) -> Self {
+        Self { inner, renamer }
+    }
+}
+
+impl Renamer for Box<dyn Fn(&str) -> String + Sync + Send> {
+    fn rename(&self, v: &str) -> std::borrow::Cow<'_, str> {
+        std::borrow::Cow::Owned(self(v))
+    }
+}

candle-nn/tests/ops.rs

@@ -4,11 +4,9 @@ extern crate intel_mkl_src;
 #[cfg(feature = "accelerate")]
 extern crate accelerate_src;
 
-use candle::{test_utils::to_vec3_round, Device, Result, Tensor};
+use candle::{test_device, test_utils::to_vec3_round, Device, Result, Tensor};
 
-#[test]
-fn softmax() -> Result<()> {
-    let device = &Device::Cpu;
+fn softmax(device: &Device) -> Result<()> {
     let data = &[[[3f32, 1., 4.], [1., 5., 9.]], [[2., 1., 7.], [8., 2., 8.]]];
     let tensor = Tensor::new(data, device)?;
     let t0 = candle_nn::ops::softmax(&tensor.log()?, 0)?;
@@ -54,6 +52,31 @@ fn softmax() -> Result<()> {
     Ok(())
 }
 
+fn rms_norm(device: &Device) -> Result<()> {
+    let data = &[[[3f32, 1., 4.], [1., 5., 9.]], [[2., 1., 7.], [8., 2., 8.]]];
+    let tensor = Tensor::new(data, device)?;
+    let alpha = Tensor::new(&[1f32, 2f32, 3f32], device)?;
+    let t = candle_nn::ops::rms_norm(&tensor, &alpha, 1e-5)?;
+    assert_eq!(
+        to_vec3_round(&t, 4)?,
+        &[
+            [[1.019, 0.6794, 4.0762], [0.1674, 1.6744, 4.521]],
+            [[0.4714, 0.4714, 4.9497], [1.206, 0.603, 3.6181]]
+        ]
+    );
+    let t2 = candle_nn::ops::rms_norm_slow(&tensor, &alpha, 1e-5)?;
+    assert_eq!(
+        to_vec3_round(&t2, 4)?,
+        &[
+            [[1.019, 0.6794, 4.0762], [0.1674, 1.6744, 4.521]],
+            [[0.4714, 0.4714, 4.9497], [1.206, 0.603, 3.6181]]
+        ]
+    );
+    let diff = (t - t2)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    assert!(diff < 1e-5);
+    Ok(())
+}
+
 #[test]
 fn softmax_numerical_stability() -> Result<()> {
     let dev = &Device::Cpu;
@@ -62,3 +85,93 @@ fn softmax_numerical_stability() -> Result<()> {
     assert_eq!(softmax.to_vec1::<f32>()?, &[1f32, 0.]);
     Ok(())
 }
+
+fn ropei(device: &Device) -> Result<()> {
+    use rand::{rngs::StdRng, Rng, SeedableRng};
+
+    let (b_size, num_head, seq_len, head_dim) = (2, 5, 10, 16);
+    let el_count = b_size * num_head * seq_len * head_dim;
+    let mut rng = StdRng::seed_from_u64(299792458);
+    let src: Vec<f32> = (0..el_count).map(|_| rng.gen::<f32>()).collect();
+    let cos: Vec<f32> = (0..seq_len * head_dim / 2)
+        .map(|_| rng.gen::<f32>())
+        .collect();
+    let sin: Vec<f32> = (0..seq_len * head_dim / 2)
+        .map(|_| rng.gen::<f32>())
+        .collect();
+    let src = Tensor::from_vec(src, (b_size, num_head, seq_len, head_dim), device)?;
+    let cos = Tensor::from_vec(cos, (seq_len, head_dim / 2), device)?;
+    let sin = Tensor::from_vec(sin, (seq_len, head_dim / 2), device)?;
+    let rope1 = candle_nn::rotary_emb::rope_i(&src, &cos, &sin)?;
+    let rope2 = candle_nn::rotary_emb::rope_i_slow(&src, &cos, &sin)?;
+    let sum_diff = (rope1 - rope2)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    if device.is_cpu() {
+        assert_eq!(sum_diff, 0.);
+    } else {
+        assert!(sum_diff < 1e-4);
+    }
+    Ok(())
+}
+
+fn rope(device: &Device) -> Result<()> {
+    use rand::{rngs::StdRng, Rng, SeedableRng};
+
+    let (b_size, num_head, seq_len, head_dim) = (2, 5, 10, 16);
+    let el_count = b_size * num_head * seq_len * head_dim;
+    let mut rng = StdRng::seed_from_u64(299792458);
+    let src: Vec<f32> = (0..el_count).map(|_| rng.gen::<f32>()).collect();
+    let cos: Vec<f32> = (0..seq_len * head_dim / 2)
+        .map(|_| rng.gen::<f32>())
+        .collect();
+    let sin: Vec<f32> = (0..seq_len * head_dim / 2)
+        .map(|_| rng.gen::<f32>())
+        .collect();
+    let src = Tensor::from_vec(src, (b_size, num_head, seq_len, head_dim), device)?;
+    let cos = Tensor::from_vec(cos, (seq_len, head_dim / 2), device)?;
+    let sin = Tensor::from_vec(sin, (seq_len, head_dim / 2), device)?;
+    let rope1 = candle_nn::rotary_emb::rope(&src, &cos, &sin)?;
+    let rope2 = candle_nn::rotary_emb::rope_slow(&src, &cos, &sin)?;
+    let sum_diff = (rope1 - rope2)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    if device.is_cpu() {
+        assert_eq!(sum_diff, 0.);
+    } else {
+        assert!(sum_diff < 1e-4);
+    }
+    Ok(())
+}
+
+fn rope_thd(device: &Device) -> Result<()> {
+    use rand::{rngs::StdRng, Rng, SeedableRng};
+
+    let (b_size, num_head, seq_len, head_dim) = (2, 5, 10, 16);
+    let el_count = b_size * num_head * seq_len * head_dim;
+    let mut rng = StdRng::seed_from_u64(299792458);
+    let src: Vec<f32> = (0..el_count).map(|_| rng.gen::<f32>()).collect();
+    let cos: Vec<f32> = (0..seq_len * head_dim / 2)
+        .map(|_| rng.gen::<f32>())
+        .collect();
+    let sin: Vec<f32> = (0..seq_len * head_dim / 2)
+        .map(|_| rng.gen::<f32>())
+        .collect();
+    let src = Tensor::from_vec(src, (b_size, num_head, seq_len, head_dim), device)?;
+    let cos = Tensor::from_vec(cos, (seq_len, head_dim / 2), device)?;
+    let sin = Tensor::from_vec(sin, (seq_len, head_dim / 2), device)?;
+    let rope1 = {
+        let src = src.transpose(1, 2)?.contiguous()?;
+        candle_nn::rotary_emb::rope_thd(&src, &cos, &sin)?.transpose(1, 2)?
+    };
+    let rope2 = candle_nn::rotary_emb::rope_slow(&src, &cos, &sin)?;
+    let sum_diff = (rope1 - rope2)?.abs()?.sum_all()?.to_vec0::<f32>()?;
+    if device.is_cpu() {
+        assert_eq!(sum_diff, 0.);
+    } else {
+        assert!(sum_diff < 1e-4);
+    }
+    Ok(())
+}
+
+test_device!(ropei, ropei_cpu, ropei_gpu, ropei_metal);
+test_device!(rope, rope_cpu, rope_gpu, rope_metal);
+test_device!(rope_thd, rope_thd_cpu, rope_thd_gpu, rope_thd_metal);
+test_device!(softmax, softmax_cpu, softmax_gpu, softmax_metal);
+test_device!(rms_norm, rms_norm_cpu, rms_norm_gpu, rms_norm_metal);