huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-20 12:06:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Metal bgemm min changes (#2364)

Browse Source 
* Add updated mfa metallib

* Add bgemm and tests
This commit is contained in:
ivarflakstad
2024-08-01 16:06:04 +08:00
committed by GitHub
parent 8696cf6494
commit fea46cb719
3 changed files with 76 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
candle-metal-kernels/src/lib.rs
View File

@ -19,6 +19,7 @@ const CAST: &str = include_str!("cast.metal");
const CONV: &str = include_str!("conv.metal");
const REDUCE: &str = include_str!("reduce.metal");
const RANDOM: &str = include_str!("random.metal");
// Current source: https://github.com/ivarflakstad/metal-flash-attention/tree/candle
const MFA: &[u8] = include_bytes!("libMetalFlashAttention.metallib");
const QUANTIZED: &str = include_str!("quantized.metal");
const SORT: &str = include_str!("sort.metal");
@ -1564,6 +1565,7 @@ pub fn call_gemm(
let bytes = match name {
"sgemm" => 4,
"hgemm" => 2,
"bgemm" => 2,
other => {
return Err(MetalKernelError::LoadLibraryError(format!(
"{other} is not a valid kernel for gemm"