huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 10:38:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: larger batches

Browse Source
This commit is contained in:
FL33TW00D
2024-01-18 14:30:14 +00:00
parent 1cf34368b7
commit 4f79f5df8a
2 changed files with 6 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
candle-metal-kernels/src/lib.rs
View File

@ -1359,13 +1359,12 @@ pub fn call_gemm(
// TODO byte_stride_d
let byte_stride_d = 0;
let mut buffer: Vec<u64> = Vec::with_capacity(b * 4);
for i in 0..b {
buffer.push((i * byte_stride_a) as u64);
buffer.push((i * byte_stride_b) as u64);
buffer.push((i * byte_stride_c) as u64);
buffer.push((i * byte_stride_d) as u64);
}
let buffer: Vec<u64> = vec![
byte_stride_a as _,
byte_stride_b as _,
byte_stride_c as _,
byte_stride_d as _,
];
encoder.set_bytes(
10,
(buffer.len() * core::mem::size_of::<u64>()) as NSUInteger,

BIN
candle-metal-kernels/src/libMetalFlashAttention.metallib
View File

Binary file not shown.