huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 03:54:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

Matmul (no batch, no strided, f32, f32 only) sort of done.

Browse Source
This commit is contained in:
Nicolas Patry
2023-11-01 17:36:51 +01:00
parent 492d164235
commit 198009453a
9 changed files with 205 additions and 96 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
candle-core/src/op.rs
View File

@ -1,5 +1,5 @@
#![allow(clippy::redundant_closure_call)]
use crate::{CpuStorage, CudaStorage, MetalStorage, Layout, Result, Shape, Tensor};
use crate::{CpuStorage, CudaStorage, Layout, MetalStorage, Result, Shape, Tensor};
use half::{bf16, f16};
use num_traits::float::Float;
@ -176,7 +176,11 @@ pub trait CustomOp1 {
/// 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)> {
fn metal_fwd(
&self,
_storage: &MetalStorage,
_layout: &Layout,
) -> Result<(MetalStorage, Shape)> {
Err(crate::Error::Metal(
format!("no cuda implementation for {}", self.name()).into(),
))