Use broadcasted scalars for const tensors.

candle-core/src/backprop.rs

@@ -99,7 +99,7 @@ impl Tensor {
     pub fn backward(&self) -> Result<GradStore> {
         let sorted_nodes = self.sorted_nodes();
         let mut grads = GradStore::new();
-        grads.insert(self, self.ones_like()?);
+        grads.insert(self, self.ones_like()?.contiguous()?);
         for node in sorted_nodes.iter() {
             if node.is_variable() {
                 continue;

candle-core/src/shape.rs

@@ -3,6 +3,8 @@ use crate::{Error, Result};
 #[derive(Clone, PartialEq, Eq)]
 pub struct Shape(Vec<usize>);
 
+pub const SCALAR: Shape = Shape(vec![]);
+
 impl std::fmt::Debug for Shape {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         write!(f, "{:?}", &self.dims())

candle-core/src/tensor.rs

@@ -115,16 +115,14 @@ fn from_storage<S: Into<Shape>>(
 }
 
 impl Tensor {
-    // TODO: Maybe this should be a broadcast rather than actually creating the full tensor.
     fn ones_impl<S: Into<Shape>>(
         shape: S,
         dtype: DType,
         device: &Device,
         is_variable: bool,
     ) -> Result<Self> {
-        let shape = shape.into();
+        let storage = device.ones(&crate::shape::SCALAR, dtype)?;
-        let storage = device.ones(&shape, dtype)?;
+        from_storage(storage, crate::shape::SCALAR, None, is_variable).broadcast_as(shape)
-        Ok(from_storage(storage, shape, None, is_variable))
     }
 
     pub fn ones<S: Into<Shape>>(shape: S, dtype: DType, device: &Device) -> Result<Self> {
@@ -132,6 +130,8 @@ impl Tensor {
     }
 
     pub fn ones_var<S: Into<Shape>>(shape: S, dtype: DType, device: &Device) -> Result<Self> {
+        // Maybe we should allocate some actual storage for vars rather than just using a
+        // broadcasted scalar?
         Self::ones_impl(shape, dtype, device, true)
     }
 
@@ -139,16 +139,14 @@ impl Tensor {
         Tensor::ones(self.shape(), self.dtype(), &self.device())
     }
 
-    // TODO: Maybe this should be a broadcast rather than actually creating the full tensor.
     fn zeros_impl<S: Into<Shape>>(
         shape: S,
         dtype: DType,
         device: &Device,
         is_variable: bool,
     ) -> Result<Self> {
-        let shape = shape.into();
+        let storage = device.zeros(&crate::shape::SCALAR, dtype)?;
-        let storage = device.zeros(&shape, dtype)?;
+        from_storage(storage, crate::shape::SCALAR, None, is_variable).broadcast_as(shape)
-        Ok(from_storage(storage, shape, None, is_variable))
     }
 
     pub fn zeros<S: Into<Shape>>(shape: S, dtype: DType, device: &Device) -> Result<Self> {
@@ -599,8 +597,7 @@ impl Tensor {
         &self.layout
     }
 
-    // TODO: Rename to `stride` once the PR that introduced the layout has been merged.
+    pub fn stride(&self) -> &[usize] {
-    pub fn stride_tmp(&self) -> &[usize] {
         self.layout.stride()
     }
 

candle-core/tests/grad_tests.rs

@@ -20,7 +20,6 @@ fn matmul_grad(device: &Device) -> Result<()> {
     let x = Tensor::var_from_slice(&data, (2, 2, 3), device)?;
     let data: Vec<_> = (0..12).map(|i| i as f32).collect();
     let y = Tensor::var_from_slice(&data, (2, 3, 2), device)?;
-
     let c = x.matmul(&y)?;
     let grads = c.backward()?;
     let grad_x = grads.get(&x).context("no grad for x")?;

candle-core/tests/tensor_tests.rs

@@ -263,12 +263,12 @@ fn matmul(device: &Device) -> Result<()> {
     let a_tt = a.t()?.contiguous()?.t()?;
     assert!(!a_tt.is_contiguous());
     assert_eq!(a.dims(), a_tt.dims());
-    assert_eq!(a_tt.stride_tmp(), &[6, 1, 2]);
+    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_tmp(), &[6, 1, 3]);
+    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);