TP sharding v2

Cargo.toml

@@ -19,8 +19,10 @@ byteorder = "1.4.3"
 clap = { version = "4.2.4", features = ["derive"] }
 # Re-enable this once 0.9.13 as been released as it would include the cublas-f16 changes
 # cudarc = { version = "0.9.13", optional = true, features = ["f16"] }
-cudarc = { git = "https://github.com/LaurentMazare/cudarc.git", branch = "cublas-bf16", features = ["f16"] }
+cudarc = { git = "https://github.com/coreylowman/cudarc.git", features = ["f16", "nccl"] }
-# TODO: Switch back to the official gemm implementation if we manage to upstream the changes.
+# TODO: Switch back to the official gemm implementation once the following are available.
+# https://github.com/sarah-ek/gemm/pull/8.
+# https://github.com/sarah-ek/gemm/pull/9.
 gemm = { git = "https://github.com/LaurentMazare/gemm.git" }
 hf-hub = "0.1.3"
 half = { version = "2.3.1", features = ["num-traits", "rand_distr"] }

candle-core/src/safetensors.rs

@@ -1,6 +1,7 @@
 use crate::{DType, Device, Error, Result, Tensor, WithDType};
+use safetensors::slice::SliceIterator;
 use safetensors::tensor as st;
-pub use safetensors::tensor::SafeTensors;
+use safetensors::tensor::{Dtype, SafeTensors};
 use std::borrow::Cow;
 
 impl From<DType> for st::Dtype {
@@ -63,15 +64,15 @@ impl Tensor {
     }
 }
 
-fn convert_<T: WithDType>(view: &st::TensorView<'_>, device: &Device) -> Result<Tensor> {
+fn convert_slice<T: WithDType>(data: &[u8], shape: &[usize], device: &Device) -> Result<Tensor> {
-    let v = view.data();
     let size_in_bytes = T::DTYPE.size_in_bytes();
-    let elem_count = v.len() / size_in_bytes;
+    let elem_count = data.len() / size_in_bytes;
-    if (v.as_ptr() as usize) % size_in_bytes == 0 {
+    if (data.as_ptr() as usize) % size_in_bytes == 0 {
         // SAFETY This is safe because we just checked that this
         // was correctly aligned.
-        let data: &[T] = unsafe { std::slice::from_raw_parts(v.as_ptr() as *const T, elem_count) };
+        let data: &[T] =
-        Tensor::from_slice(data, view.shape(), device)
+            unsafe { std::slice::from_raw_parts(data.as_ptr() as *const T, elem_count) };
+        Tensor::from_slice(data, shape, device)
     } else {
         // XXX: We need to specify `T` here, otherwise the compiler will infer u8 because of the following cast
         // Making this vector too small to fit a full f16/f32/f64 weights, resulting in out-of-bounds access
@@ -81,13 +82,17 @@ fn convert_<T: WithDType>(view: &st::TensorView<'_>, device: &Device) -> Result<
         // We're downgrading the `c` pointer from T to u8, which removes alignment
         // constraints.
         unsafe {
-            std::ptr::copy_nonoverlapping(v.as_ptr(), c.as_mut_ptr() as *mut u8, v.len());
+            std::ptr::copy_nonoverlapping(data.as_ptr(), c.as_mut_ptr() as *mut u8, data.len());
             c.set_len(elem_count)
         }
-        Tensor::from_slice(&c, view.shape(), device)
+        Tensor::from_slice(&c, shape, device)
     }
 }
 
+fn convert_<T: WithDType>(view: &st::TensorView<'_>, device: &Device) -> Result<Tensor> {
+    convert_slice::<T>(view.data(), view.shape(), device)
+}
+
 fn convert_back_<T: WithDType>(mut vs: Vec<T>) -> Vec<u8> {
     let size_in_bytes = T::DTYPE.size_in_bytes();
     let length = vs.len() * size_in_bytes;
@@ -112,6 +117,26 @@ impl<'a> Load for st::TensorView<'a> {
     }
 }
 
+impl Tensor {
+    pub fn from_safetensors_slice(
+        iterator: SliceIterator,
+        dtype: Dtype,
+        shape: &[usize],
+        device: &Device,
+    ) -> Result<Self> {
+        let data: Vec<u8> = iterator.into_iter().flatten().cloned().collect();
+        match dtype {
+            st::Dtype::U8 => convert_slice::<u8>(&data, shape, device),
+            st::Dtype::U32 => convert_slice::<u8>(&data, shape, device),
+            st::Dtype::BF16 => convert_slice::<half::bf16>(&data, shape, device),
+            st::Dtype::F16 => convert_slice::<half::f16>(&data, shape, device),
+            st::Dtype::F32 => convert_slice::<f32>(&data, shape, device),
+            st::Dtype::F64 => convert_slice::<f64>(&data, shape, device),
+            dtype => Err(Error::UnsupportedSafeTensorDtype(dtype)),
+        }
+    }
+}
+
 pub fn convert(view: &st::TensorView<'_>, device: &Device) -> Result<Tensor> {
     match view.dtype() {
         st::Dtype::U8 => convert_::<u8>(view, device),

candle-examples/Cargo.toml

@@ -19,6 +19,8 @@ serde = { workspace = true }
 serde_json = { workspace = true }
 num-traits = { workspace = true }
 intel-mkl-src = { workspace = true, optional = true }
+cudarc = { workspace = true, optional = true }
+half = { workspace = true, optional = true }
 
 [dev-dependencies]
 anyhow = { workspace = true }
@@ -40,3 +42,8 @@ default = []
 cuda = ["candle/cuda", "candle-nn/cuda", "candle-transformers/cuda"]
 flash-attn = ["cuda", "dep:candle-flash-attn"]
 mkl = ["dep:intel-mkl-src", "candle/mkl", "candle-nn/mkl", "candle-transformers/mkl"]
+nccl = ["dep:cudarc", "dep:half"]
+
+[[example]]
+name = "llama_multiprocess"
+required-features = ["cuda", "nccl"]

candle-examples/examples/llama_multiprocess/main.rs

@@ -0,0 +1,266 @@
+// An implementation of LLaMA https://github.com/facebookresearch/llama
+//
+// This is based on nanoGPT in a similar way to:
+// https://github.com/Lightning-AI/lit-llama/blob/main/lit_llama/model.py
+//
+// The tokenizer config can be retrieved from:
+// https://huggingface.co/hf-internal-testing/llama-tokenizer/raw/main/tokenizer.json
+//
+// In order to convert the llama weights to a .npz file, run:
+// python examples/llama/convert_checkpoint.py ..../LLaMA/7B/consolidated.00.pth
+
+#[cfg(feature = "mkl")]
+extern crate intel_mkl_src;
+
+use anyhow::{Error as E, Result};
+use clap::Parser;
+
+use candle::{DType, Device, Tensor};
+use candle_nn::VarBuilder;
+use candle_transformers::generation::LogitsProcessor;
+use cudarc::driver::safe::CudaDevice;
+use cudarc::nccl::safe::{Comm, Id};
+use hf_hub::{api::sync::Api, Repo, RepoType};
+use std::io::Write;
+use std::rc::Rc;
+
+mod model;
+use model::{Config, Llama};
+
+const MAX_SEQ_LEN: usize = 4096;
+const DEFAULT_PROMPT: &str = r"
+EDWARD:
+I wonder how our princely father 'scaped,
+Or whether he be 'scaped away or no
+From Clifford's and Northumberland's pursuit:
+Had he been ta'en, we should have heard the news;
+Had he been slain, we should have heard the news;
+Or had he 'scaped, methinks we should have heard
+The happy tidings of his good escape.
+How fares my brother? why is he so sad?
+
+RICHARD:
+I cannot joy, until I be resolved
+Where our right valiant father is become.
+I saw him in the battle range about;
+And watch'd him how he singled Clifford forth.
+Methought he bore him in the thickest troop
+As doth a lion in a herd of neat;
+Or as a bear, encompass'd round with dogs,
+Who having pinch'd a few and made them cry,
+The rest stand all aloof, and bark at him.
+So fared our father with his enemies;
+So fled his enemies my warlike father:
+Methinks, 'tis prize enough to be his son.
+See how the morning opes her golden gates,
+And takes her farewell of the glorious sun!
+How well resembles it the prime of youth,
+Trimm'd like a younker prancing to his love!
+
+EDWARD:
+Dazzle mine eyes, or do I see three suns?
+
+RICHARD:
+Three glorious suns, each one a perfect sun;
+Not separated with the racking clouds,
+But sever'd in a pale clear-shining sky.
+See, see! they join, embrace, and seem to kiss,
+As if they vow'd some league inviolable:
+Now are they but one lamp, one light, one sun.
+In this the heaven figures some event.
+
+EDWARD:
+'Tis wondrous strange, the like yet never heard of.
+I think it cites us, brother, to the field,
+That we, the sons of brave Plantagenet,
+Each one already blazing by our meeds,
+Should notwithstanding join our lights together
+And over-shine the earth as this the world.
+Whate'er it bodes, henceforward will I bear
+Upon my target three fair-shining suns.
+";
+
+#[derive(Parser, Debug)]
+#[command(author, version, about, long_about = None)]
+struct Args {
+    /// Run on CPU rather than on GPU.
+    #[arg(long)]
+    cpu: bool,
+
+    #[arg(long)]
+    num_shards: usize,
+
+    #[arg(long)]
+    rank: Option<usize>,
+
+    /// The temperature used to generate samples.
+    #[arg(long)]
+    temperature: Option<f64>,
+
+    /// 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, default_value_t = 100)]
+    sample_len: usize,
+
+    /// Disable the key-value cache.
+    #[arg(long)]
+    no_kv_cache: bool,
+
+    /// The initial prompt.
+    #[arg(long)]
+    prompt: Option<String>,
+
+    /// Use f32 computations rather than f16.
+    #[arg(long)]
+    use_f32: bool,
+
+    #[arg(long)]
+    model_id: Option<String>,
+
+    #[arg(long)]
+    v2: bool,
+}
+
+fn main() -> Result<()> {
+    use tokenizers::Tokenizer;
+
+    let args = Args::parse();
+
+    let config = Config::config_7b();
+    let dtype = if args.use_f32 { DType::F32 } else { DType::F16 };
+
+    let api = Api::new()?;
+
+    let model_id = args.model_id.unwrap_or_else(|| {
+        if args.v2 {
+            "meta-llama/Llama-2-7b-hf".to_string()
+        } else {
+            "Narsil/amall-7b".to_string()
+        }
+    });
+    println!("loading the model weights from {model_id}");
+    let repo = Repo::new(model_id, RepoType::Model);
+    let tokenizer_filename = api.get(&repo, "tokenizer.json")?;
+    let mut filenames = vec![];
+    for rfilename in [
+        "model-00001-of-00002.safetensors",
+        "model-00002-of-00002.safetensors",
+    ] {
+        let filename = api.get(&repo, rfilename)?;
+        filenames.push(filename);
+    }
+
+    if args.rank.is_none() {
+        let children: Vec<_> = (0..args.num_shards)
+            .map(|rank| {
+                let mut args: std::collections::VecDeque<_> = std::env::args().collect();
+                args.push_back("--rank".to_string());
+                args.push_back(format!("{rank}"));
+                let name = args.pop_front().unwrap();
+                std::process::Command::new(name).args(args).spawn().unwrap()
+            })
+            .collect();
+        for mut child in children {
+            child.wait().unwrap();
+        }
+        return Ok(());
+    }
+
+    let i = args.rank.unwrap();
+    let num_shards = args.num_shards;
+    let rank = i;
+    // Primitive IPC
+    let id = if rank == 0 {
+        let id = Id::new().unwrap();
+        std::fs::File::create("nccl_id.txt.tmp")?
+            .write_all(&id.internal().iter().map(|&i| i as u8).collect::<Vec<_>>())
+            .unwrap();
+        std::fs::rename("nccl_id.txt.tmp", "nccl_id.txt")?;
+        id
+    } else {
+        let path = std::path::PathBuf::from("nccl_id.txt");
+        while !path.exists() {
+            std::thread::sleep(std::time::Duration::from_secs(1));
+        }
+        let data = std::fs::read("nccl_id.txt")?;
+        let internal: [i8; 128] = data
+            .into_iter()
+            .map(|i| i as i8)
+            .collect::<Vec<_>>()
+            .try_into()
+            .unwrap();
+        let id: Id = Id::uninit(internal);
+        id
+    };
+    let device = CudaDevice::new(i)?;
+    let comm = Rc::new(Comm::from_rank(device, i, num_shards, id).unwrap());
+    if rank == 0 {
+        std::fs::remove_file("nccl_id.txt")?;
+    }
+    println!("Rank {rank:?} spawned");
+
+    let device = Device::new_cuda(i)?;
+    let cache = model::Cache::new(!args.no_kv_cache, &config, &device)?;
+
+    println!("building the model");
+    let handles = filenames
+        .iter()
+        .map(|f| Ok(unsafe { candle::safetensors::MmapedFile::new(f.as_path())? }))
+        .collect::<Result<Vec<_>>>()?;
+    let tensors: Vec<_> = handles
+        .iter()
+        .map(|h| Ok(h.deserialize()?))
+        .collect::<Result<Vec<_>>>()?;
+
+    let vb = VarBuilder::from_safetensors(tensors, dtype, &device);
+    let llama = Llama::load(vb, &cache, &config, comm)?;
+    let tokenizer = Tokenizer::from_file(tokenizer_filename).map_err(E::msg)?;
+
+    let prompt = args.prompt.as_ref().map_or(DEFAULT_PROMPT, |p| p.as_str());
+    let mut tokens = tokenizer
+        .encode(prompt, true)
+        .map_err(E::msg)?
+        .get_ids()
+        .to_vec();
+
+    println!("starting the inference loop");
+    let mut logits_processor = LogitsProcessor::new(args.seed, args.temperature);
+    let mut new_tokens = vec![];
+    let start_gen = std::time::Instant::now();
+    let mut index_pos = 0;
+    for index in 0..args.sample_len {
+        let start_gen = std::time::Instant::now();
+        let context_size = if cache.use_kv_cache && index > 0 {
+            1
+        } else {
+            tokens.len()
+        };
+        let ctxt = &tokens[tokens.len().saturating_sub(context_size)..];
+        let input = Tensor::new(ctxt, &device)?.unsqueeze(0)?;
+        let logits = llama.forward(&input, index_pos)?;
+        let logits = logits.squeeze(0)?;
+        index_pos += ctxt.len();
+
+        let next_token = logits_processor.sample(&logits)?;
+        tokens.push(next_token);
+        new_tokens.push(next_token);
+        println!("> {:?}", start_gen.elapsed());
+        println!(
+            "{} token: {} '{}'",
+            index + 1,
+            next_token,
+            tokenizer.decode(vec![next_token], true).map_err(E::msg)?
+        );
+    }
+    let dt = start_gen.elapsed();
+    println!(
+        "{} tokens generated ({} token/s)\n----\n{}\n----",
+        args.sample_len,
+        args.sample_len as f64 / dt.as_secs_f64(),
+        tokenizer.decode(new_tokens, true).map_err(E::msg)?
+    );
+    Ok(())
+}

candle-examples/examples/llama_multiprocess/model.rs

@@ -0,0 +1,465 @@
+use candle::{DType, Device, IndexOp, Result, Tensor, D};
+use candle_nn::{Embedding, Linear, VarBuilder};
+use cudarc::driver::safe::CudaSlice;
+use cudarc::nccl::safe::{Comm, ReduceOp};
+use half::f16;
+use std::collections::HashMap;
+use std::rc::Rc;
+use std::sync::{Arc, Mutex};
+
+use super::MAX_SEQ_LEN;
+
+struct TensorParallelColumnLinear {
+    linear: Linear,
+}
+
+impl TensorParallelColumnLinear {
+    fn new(linear: Linear) -> Self {
+        Self { linear }
+    }
+    fn forward(&self, x: &Tensor) -> Result<Tensor> {
+        self.linear.forward(x)
+    }
+}
+
+struct TensorParallelRowLinear {
+    linear: Linear,
+    comm: Rc<Comm>,
+}
+
+fn all_reduce_sum(x: &Tensor, comm: &Rc<Comm>) -> Result<Tensor> {
+    Ok(x.clone())
+    // let n = x.shape().elem_count();
+    // let cuda_slice: CudaSlice<f16> = x.try_into()?;
+    // let dev = cuda_slice.device();
+    // let mut slice_receive = dev.alloc_zeros(n).unwrap();
+    // comm.all_reduce(cuda_slice, &mut slice_receive, &ReduceOp::Sum).unwrap();
+    // Tensor::from_raw_storage(slice_receive, x.shape())
+}
+
+impl TensorParallelRowLinear {
+    fn new(linear: Linear, comm: Rc<Comm>) -> Self {
+        Self { linear, comm }
+    }
+    fn forward(&self, x: &Tensor) -> Result<Tensor> {
+        let x = self.linear.forward(x)?;
+        all_reduce_sum(&x, &self.comm)
+    }
+}
+
+impl TensorParallelColumnLinear {
+    fn load(vb: VarBuilder, comm: Rc<Comm>) -> Result<Self> {
+        let rank = comm.rank();
+        let size = comm.world_size();
+        let weight = vb.get_sharded("weight", 0, rank, size)?;
+        Ok(Self::new(Linear::new(weight, None)))
+    }
+
+    fn load_multi(vb: VarBuilder, prefixes: &[&str], comm: Rc<Comm>) -> Result<Self> {
+        let rank = comm.rank();
+        let size = comm.world_size();
+        let weights: Vec<_> = prefixes
+            .iter()
+            .map(|p| vb.pp(p).get_sharded("weight", 0, rank, size).unwrap())
+            .collect();
+        let weight = Tensor::cat(&weights, 0)?;
+        Ok(Self::new(Linear::new(weight, None)))
+    }
+}
+
+impl TensorParallelRowLinear {
+    fn load(vb: VarBuilder, comm: Rc<Comm>) -> Result<Self> {
+        let rank = comm.rank();
+        let size = comm.world_size();
+        let weight = vb.get_sharded("weight", 1, rank, size)?;
+        Ok(Self::new(Linear::new(weight, None), comm.clone()))
+    }
+}
+
+pub struct Config {
+    pub hidden_size: usize,
+    pub intermediate_size: usize,
+    pub vocab_size: usize,
+    pub n_layer: usize,
+    pub n_head: usize,
+    pub n_embd: usize,
+    pub n_key_value_head: usize,
+}
+
+impl Config {
+    pub fn config_7b() -> Self {
+        Self {
+            hidden_size: 4096,
+            intermediate_size: 11008,
+            vocab_size: 32000,
+            n_layer: 32,
+            n_head: 32,
+            n_embd: 4096,
+            n_key_value_head: 32,
+        }
+    }
+}
+
+#[derive(Clone)]
+pub struct Cache {
+    masks: Arc<Mutex<HashMap<usize, Tensor>>>,
+    pub use_kv_cache: bool,
+    #[allow(clippy::type_complexity)]
+    kvs: Arc<Mutex<Vec<Option<(Tensor, Tensor)>>>>,
+    cos: Tensor,
+    sin: Tensor,
+    device: Device,
+}
+
+impl Cache {
+    pub fn new(use_kv_cache: bool, config: &Config, device: &Device) -> Result<Self> {
+        // precompute freqs_cis
+        let n_elem = config.n_embd / config.n_head;
+        let theta: Vec<_> = (0..n_elem)
+            .step_by(2)
+            .map(|i| 1f32 / 10000f32.powf(i as f32 / n_elem as f32))
+            .collect();
+        let theta = Tensor::new(theta.as_slice(), device)?;
+        let idx_theta = Tensor::arange(0, MAX_SEQ_LEN as u32, device)?
+            .to_dtype(DType::F32)?
+            .reshape((MAX_SEQ_LEN, 1))?
+            .matmul(&theta.reshape((1, theta.elem_count()))?)?;
+        // This is different from the paper, see:
+        // https://github.com/huggingface/transformers/blob/6112b1c6442aaf7affd2b0676a1cd4eee30c45cf/src/transformers/models/llama/modeling_llama.py#L112
+        let idx_theta = Tensor::cat(&[&idx_theta, &idx_theta], D::Minus1)?;
+        let cos = idx_theta.cos()?;
+        let sin = idx_theta.sin()?;
+        Ok(Self {
+            masks: Arc::new(Mutex::new(HashMap::new())),
+            use_kv_cache,
+            kvs: Arc::new(Mutex::new(vec![None; config.n_layer])),
+            device: device.clone(),
+            cos,
+            sin,
+        })
+    }
+
+    fn mask(&self, t: usize) -> Result<Tensor> {
+        let mut masks = self.masks.lock().unwrap();
+        if let Some(mask) = masks.get(&t) {
+            Ok(mask.clone())
+        } else {
+            // TODO: If we support bool or u8 tensors, this would be better.
+            let mask: Vec<_> = (0..t)
+                .flat_map(|i| (0..t).map(move |j| u32::from(j > i)))
+                .collect();
+            let mask = Tensor::from_slice(&mask, (t, t), &self.device)?;
+            masks.insert(t, mask.clone());
+            Ok(mask)
+        }
+    }
+}
+
+fn silu(xs: &Tensor) -> Result<Tensor> {
+    xs / (xs.neg()?.exp()? + 1.0)?
+}
+
+fn linear(size1: usize, size2: usize, vb: VarBuilder) -> Result<Linear> {
+    let weight = vb.get((size2, size1), "weight")?;
+    Ok(Linear::new(weight, None))
+}
+
+fn embedding(cfg: &Config, vb: VarBuilder) -> Result<Embedding> {
+    let embeddings = vb.get((cfg.vocab_size, cfg.hidden_size), "weight")?;
+    Ok(Embedding::new(embeddings, cfg.hidden_size))
+}
+
+struct RmsNorm {
+    scale: Tensor,
+}
+
+impl RmsNorm {
+    fn load(size: usize, vb: VarBuilder) -> Result<Self> {
+        let scale = vb.get(size, "weight")?;
+        Ok(Self::new(scale))
+    }
+
+    fn new(scale: Tensor) -> Self {
+        Self { scale }
+    }
+
+    fn forward(&self, x: &Tensor) -> Result<Tensor> {
+        let in_dtype = x.dtype();
+        // This is a no-op if x's dtype is already f32.
+        let x = x.to_dtype(DType::F32)?;
+        let (b_sz, seq_len, hidden_size) = x.shape().r3()?;
+        let norm_x = (x.sqr()?.sum_keepdim(2)? / hidden_size as f64)?;
+        let norm_x = norm_x.broadcast_as((b_sz, seq_len, hidden_size))?;
+        let x_normed = (x / (norm_x + 1e-6)?.sqrt()?)?;
+        let size = self.scale.shape().r1()?;
+        let scale = self
+            .scale
+            .to_dtype(DType::F32)?
+            .broadcast_as((b_sz, seq_len, size))?;
+        let x = (scale * x_normed)?;
+        let x = x.to_dtype(in_dtype)?;
+        Ok(x)
+    }
+}
+
+struct CausalSelfAttention {
+    qkv_proj: TensorParallelColumnLinear,
+    o_proj: TensorParallelRowLinear,
+    n_head: usize,
+    n_key_value_head: usize,
+    head_dim: usize,
+    cache: Cache,
+}
+
+impl CausalSelfAttention {
+    fn apply_rotary_emb(&self, x: &Tensor, index_pos: usize) -> Result<Tensor> {
+        let (b_sz, _, seq_len, n_embd) = x.shape().r4()?;
+        let cos = self.cache.cos.narrow(0, index_pos, seq_len)?;
+        let sin = self.cache.sin.narrow(0, index_pos, seq_len)?;
+        let cos = cos.broadcast_as((b_sz, 1, seq_len, n_embd))?;
+        let sin = sin.broadcast_as((b_sz, 1, seq_len, n_embd))?;
+        let x1 = x.narrow(D::Minus1, 0, n_embd / 2)?;
+        let x2 = x.narrow(D::Minus1, n_embd / 2, n_embd / 2)?;
+        let rotate_x = Tensor::cat(&[&x2.neg()?, &x1], D::Minus1)?;
+        let rope = (x.broadcast_mul(&cos)? + rotate_x.broadcast_mul(&sin)?)?;
+        Ok(rope)
+    }
+
+    fn forward(&self, x: &Tensor, index_pos: usize, block_idx: usize) -> Result<Tensor> {
+        let x_dtype = x.dtype();
+        let (b_sz, seq_len, _) = x.shape().r3()?;
+
+        let qkv = self.qkv_proj.forward(x)?;
+        let n_embd = self.n_head * self.head_dim;
+
+        let q = qkv.i((.., .., ..self.n_head * self.head_dim))?;
+        let k = qkv.i((
+            ..,
+            ..,
+            self.n_head * self.head_dim
+                ..self.n_head * self.head_dim + self.n_key_value_head * self.head_dim,
+        ))?;
+        let v = qkv.i((
+            ..,
+            ..,
+            self.n_head * self.head_dim + self.n_key_value_head * self.head_dim..,
+        ))?;
+        // todo!("Q {:?} K {:?} V {:?} - x {:?}", q.shape(), k.shape(), v.shape(), x.shape());
+
+        let q = q
+            .reshape((b_sz, seq_len, self.n_head, self.head_dim))?
+            .transpose(1, 2)?
+            .to_dtype(DType::F32)?;
+        let k = k
+            .reshape((b_sz, seq_len, self.n_key_value_head, self.head_dim))?
+            .transpose(1, 2)?
+            .to_dtype(DType::F32)?;
+        let mut v = v
+            .reshape((b_sz, seq_len, self.n_key_value_head, self.head_dim))?
+            .transpose(1, 2)?
+            .to_dtype(DType::F32)?;
+
+        let q = self.apply_rotary_emb(&q, index_pos)?;
+        let mut k = self.apply_rotary_emb(&k, index_pos)?;
+
+        if self.cache.use_kv_cache {
+            let mut cache = self.cache.kvs.lock().unwrap();
+            if let Some((cache_k, cache_v)) = &cache[block_idx] {
+                k = Tensor::cat(&[cache_k, &k], 2)?.contiguous()?;
+                v = Tensor::cat(&[cache_v, &v], 2)?.contiguous()?;
+                let k_seq_len = k.dims()[1];
+                if k_seq_len > MAX_SEQ_LEN {
+                    k = k
+                        .narrow(D::Minus1, k_seq_len - MAX_SEQ_LEN, MAX_SEQ_LEN)?
+                        .contiguous()?
+                }
+                let v_seq_len = v.dims()[1];
+                if v_seq_len > 2 * MAX_SEQ_LEN {
+                    v = v
+                        .narrow(D::Minus1, v_seq_len - MAX_SEQ_LEN, MAX_SEQ_LEN)?
+                        .contiguous()?
+                }
+            }
+            cache[block_idx] = Some((k.clone(), v.clone()))
+        }
+
+        let k = self.repeat_kv(k)?;
+        let v = self.repeat_kv(v)?;
+        let att = (q.matmul(&k.t()?)? / (self.head_dim as f64).sqrt())?;
+        let mask = self.cache.mask(seq_len)?.broadcast_as(att.shape())?;
+        let att = masked_fill(&att, &mask, f32::NEG_INFINITY)?;
+        let att = att.softmax(D::Minus1)?;
+        // Convert to contiguous as matmul doesn't support strided vs for now.
+        let y = att.matmul(&v.contiguous()?)?;
+        let y = y.transpose(1, 2)?.reshape(&[b_sz, seq_len, n_embd])?;
+        let y = y.to_dtype(x_dtype)?;
+        let y = self.o_proj.forward(&y)?;
+        Ok(y)
+    }
+
+    fn repeat_kv(&self, x: Tensor) -> Result<Tensor> {
+        let n_rep = self.n_head / self.n_key_value_head;
+        if n_rep == 1 {
+            Ok(x)
+        } else {
+            let (b_sz, n_kv_head, seq_len, head_dim) = x.shape().r4()?;
+            let x = x
+                .unsqueeze(2)?
+                .expand((b_sz, n_kv_head, n_rep, seq_len, head_dim))?
+                .reshape((b_sz, n_kv_head, n_rep, seq_len, head_dim))?;
+            Ok(x)
+        }
+    }
+
+    fn load(vb: VarBuilder, cache: &Cache, cfg: &Config, comm: Rc<Comm>) -> Result<Self> {
+        let size_in = cfg.hidden_size;
+        let size_q = (cfg.hidden_size / cfg.n_head) * cfg.n_head;
+        let size_kv = (cfg.hidden_size / cfg.n_head) * cfg.n_key_value_head;
+
+        let qkv_proj = TensorParallelColumnLinear::load_multi(
+            vb.clone(),
+            &["q_proj", "k_proj", "v_proj"],
+            comm.clone(),
+        )?;
+        let o_proj = TensorParallelRowLinear::load(vb.pp("o_proj"), comm.clone())?;
+        Ok(Self {
+            qkv_proj,
+            o_proj,
+            n_head: cfg.n_head / comm.world_size(),
+            n_key_value_head: cfg.n_key_value_head / comm.world_size(),
+            head_dim: cfg.hidden_size / cfg.n_head,
+            cache: cache.clone(),
+        })
+    }
+}
+
+fn masked_fill(on_false: &Tensor, mask: &Tensor, on_true: f32) -> Result<Tensor> {
+    let shape = mask.shape();
+    let on_true = Tensor::new(on_true, on_false.device())?.broadcast_as(shape.dims())?;
+    let m = mask.where_cond(&on_true, on_false)?;
+    Ok(m)
+}
+
+struct Mlp {
+    c_fc1: TensorParallelColumnLinear,
+    c_fc2: TensorParallelColumnLinear,
+    c_proj: TensorParallelRowLinear,
+}
+
+impl Mlp {
+    fn new(
+        c_fc1: TensorParallelColumnLinear,
+        c_fc2: TensorParallelColumnLinear,
+        c_proj: TensorParallelRowLinear,
+    ) -> Self {
+        Self {
+            c_fc1,
+            c_fc2,
+            c_proj,
+        }
+    }
+
+    fn forward(&self, x: &Tensor) -> Result<Tensor> {
+        let x = (silu(&self.c_fc1.forward(x)?)? * self.c_fc2.forward(x)?)?;
+        self.c_proj.forward(&x)
+    }
+
+    fn load(vb: VarBuilder, cfg: &Config, comm: Rc<Comm>) -> Result<Self> {
+        let h_size = cfg.hidden_size;
+        let i_size = cfg.intermediate_size;
+        let c_fc1 = TensorParallelColumnLinear::load(vb.pp("gate_proj"), comm.clone())?;
+        let c_fc2 = TensorParallelColumnLinear::load(vb.pp("up_proj"), comm.clone())?;
+        let c_proj = TensorParallelRowLinear::load(vb.pp("down_proj"), comm.clone())?;
+        Ok(Self::new(c_fc1, c_fc2, c_proj))
+    }
+}
+
+struct Block {
+    rms_1: RmsNorm,
+    attn: CausalSelfAttention,
+    rms_2: RmsNorm,
+    mlp: Mlp,
+}
+
+impl Block {
+    fn new(rms_1: RmsNorm, attn: CausalSelfAttention, rms_2: RmsNorm, mlp: Mlp) -> Self {
+        Self {
+            rms_1,
+            attn,
+            rms_2,
+            mlp,
+        }
+    }
+
+    fn forward(&self, x: &Tensor, index_pos: usize, block_idx: usize) -> Result<Tensor> {
+        let residual = x;
+        let x = self.rms_1.forward(x)?;
+        let x = (self.attn.forward(&x, index_pos, block_idx)? + residual)?;
+        let residual = &x;
+        let x = (self.mlp.forward(&self.rms_2.forward(&x)?)? + residual)?;
+        Ok(x)
+    }
+
+    fn load(vb: VarBuilder, cache: &Cache, cfg: &Config, comm: Rc<Comm>) -> Result<Self> {
+        let attn = CausalSelfAttention::load(vb.pp("self_attn"), cache, cfg, comm.clone())?;
+        let mlp = Mlp::load(vb.pp("mlp"), cfg, comm.clone())?;
+        let input_layernorm = RmsNorm::load(cfg.hidden_size, vb.pp("input_layernorm"))?;
+        let post_attention_layernorm =
+            RmsNorm::load(cfg.hidden_size, vb.pp("post_attention_layernorm"))?;
+        Ok(Self::new(
+            input_layernorm,
+            attn,
+            post_attention_layernorm,
+            mlp,
+        ))
+    }
+}
+
+pub struct Llama {
+    wte: Embedding,
+    blocks: Vec<Block>,
+    ln_f: RmsNorm,
+    lm_head: Linear,
+}
+
+impl Llama {
+    fn new(wte: Embedding, blocks: Vec<Block>, ln_f: RmsNorm, lm_head: Linear) -> Self {
+        Self {
+            wte,
+            blocks,
+            ln_f,
+            lm_head,
+        }
+    }
+
+    pub fn forward(&self, x: &Tensor, index_pos: usize) -> Result<Tensor> {
+        let (_b_sz, seq_len) = x.shape().r2()?;
+        let mut x = self.wte.forward(x)?;
+        for (block_idx, block) in self.blocks.iter().enumerate() {
+            x = block.forward(&x, index_pos, block_idx)?;
+        }
+        let x = self.ln_f.forward(&x)?;
+        let x = x.i((.., seq_len - 1, ..))?;
+        let logits = self.lm_head.forward(&x)?;
+        logits.to_dtype(DType::F32)
+    }
+
+    pub fn load(vb: VarBuilder, cache: &Cache, cfg: &Config, comm: Rc<Comm>) -> Result<Self> {
+        let wte = embedding(cfg, vb.pp("model.embed_tokens"))?;
+        let lm_head = linear(cfg.hidden_size, cfg.vocab_size, vb.pp("lm_head"))?;
+        let norm = RmsNorm::load(cfg.hidden_size, vb.pp("model.norm"))?;
+        let blocks: Vec<_> = (0..cfg.n_layer)
+            .map(|i| {
+                Block::load(
+                    vb.pp(&format!("model.layers.{i}")),
+                    cache,
+                    cfg,
+                    comm.clone(),
+                )
+                .unwrap()
+            })
+            .collect();
+
+        Ok(Self::new(wte, blocks, norm, lm_head))
+    }
+}

candle-nn/Cargo.toml

@@ -14,6 +14,7 @@ readme = "README.md"
 candle = { path = "../candle-core" }
 thiserror = { workspace = true }
 intel-mkl-src = { workspace = true, optional = true }
+safetensors = { workspace = true }
 
 [dev-dependencies]
 anyhow = { workspace = true }

candle-nn/src/var_builder.rs

@@ -1,7 +1,6 @@
-use candle::{
+use candle::{safetensors::Load, DType, Device, Error, Result, Shape, Tensor};
-    safetensors::{Load, SafeTensors},
+use safetensors::slice::IndexOp;
-    DType, Device, Error, Result, Shape, Tensor,
+use safetensors::tensor::SafeTensors;
-};
 use std::collections::HashMap;
 use std::sync::Arc;
 
@@ -71,7 +70,7 @@ impl<'a> TensorData<'a> {
 #[derive(Clone)]
 pub struct VarBuilder<'a> {
     data: Arc<TensorData<'a>>,
-    path: Vec<String>,
+    pub path: Vec<String>,
 }
 
 impl<'a> VarBuilder<'a> {
@@ -137,6 +136,55 @@ impl<'a> VarBuilder<'a> {
 }
 
 impl<'a> VarBuilder<'a> {
+    pub fn get_sharded(
+        &self,
+        tensor_name: &str,
+        dim: usize,
+        rank: usize,
+        world_size: usize,
+    ) -> Result<Tensor> {
+        let data = self.data.as_ref();
+        let path = if self.path.is_empty() {
+            tensor_name.to_string()
+        } else {
+            [&self.path.join("."), tensor_name].join(".")
+        };
+        let tensor = match &self.data.tensors {
+            Tensors::SafeTensorWithRouting {
+                routing,
+                safetensors,
+            } => {
+                let index = routing.get(&path).ok_or_else(|| {
+                    Error::CannotFindTensor {
+                        path: path.to_string(),
+                    }
+                    .bt()
+                })?;
+
+                let view = safetensors[*index].tensor(&path)?;
+                let dtype = view.dtype();
+                let mut shape = view.shape().to_vec();
+                let size = shape[dim];
+                let block_size = size / world_size;
+                let start = rank * block_size;
+                let stop = (rank + 1) * block_size;
+
+                let iterator = if dim == 0 {
+                    view.slice(start..stop).unwrap()
+                } else if dim == 1 {
+                    view.slice((.., start..stop)).unwrap()
+                } else {
+                    unimplemented!("Get sharded on dimensions != 0 or 1");
+                };
+
+                shape[dim] = block_size;
+
+                Tensor::from_safetensors_slice(iterator, dtype, &shape, &data.device)?
+            }
+            _ => unimplemented!(),
+        };
+        Ok(tensor)
+    }
     pub fn get<S: Into<Shape>>(&self, s: S, tensor_name: &str) -> Result<Tensor> {
         let data = self.data.as_ref();
         let s: Shape = s.into();

candle-wasm-examples/whisper/Cargo.toml

@@ -15,6 +15,7 @@ candle = { path = "../../candle-core" }
 candle-nn = { path = "../../candle-nn" }
 num-traits = { workspace = true }
 tokenizers = { workspace = true, features = ["unstable_wasm"] }
+safetensors = { workspace = true }
 
 # App crates.
 anyhow = { workspace = true }

candle-wasm-examples/whisper/src/worker.rs

@@ -236,11 +236,11 @@ impl Decoder {
         let device = Device::Cpu;
         let tokenizer = Tokenizer::from_bytes(&md.tokenizer).map_err(anyhow::Error::msg)?;
 
-        let mel_filters = candle::safetensors::SafeTensors::deserialize(&md.mel_filters)?;
+        let mel_filters = safetensors::tensor::SafeTensors::deserialize(&md.mel_filters)?;
         let mel_filters = mel_filters.tensor("mel_80")?.load(&device)?;
         console_log!("loaded mel filters {:?}", mel_filters.shape());
         let mel_filters = mel_filters.flatten_all()?.to_vec1::<f32>()?;
-        let weights = candle::safetensors::SafeTensors::deserialize(&md.weights)?;
+        let weights = safetensors::tensor::SafeTensors::deserialize(&md.weights)?;
         let vb = VarBuilder::from_safetensors(vec![weights], DTYPE, &device);
         let config = Config::tiny_en();
         let whisper = Whisper::load(&vb, config)?;