[Bugfix][CI] fix typos (#34934)
Signed-off-by: 1195343015 <1195343015@qq.com> Signed-off-by: Jiayi Yan <66017932+1195343015@users.noreply.github.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
co-authored by
Harry Mellor
parent
8c760b6ab6
commit
6a895197fa
@@ -255,7 +255,7 @@ def _flashinfer_fp8_blockscale_gemm_impl(
|
||||
|
||||
This batch-size-dependent selection is essential for maintaining model accuracy.
|
||||
Benchmarks on GSM8K show a significant accuracy gap (88% vs 95%) for DeepSeek-V3.1
|
||||
when using FlashInfer's DeepGEMM on M>=32. The M < 32 strategy fixes the accurracy
|
||||
when using FlashInfer's DeepGEMM on M>=32. The M < 32 strategy fixes the accuracy
|
||||
drop.
|
||||
|
||||
Args:
|
||||
|
||||
@@ -39,7 +39,7 @@ def query_machete_supported_group_sizes(act_type: torch.dtype) -> list[int]:
|
||||
|
||||
|
||||
def check_machete_supports_shape(
|
||||
in_features: int, out_featrues: int
|
||||
in_features: int, out_features: int
|
||||
) -> tuple[bool, str | None]:
|
||||
if in_features % MACHETE_PREPACKED_BLOCK_SHAPE[0] != 0:
|
||||
return (
|
||||
@@ -47,7 +47,7 @@ def check_machete_supports_shape(
|
||||
"Input features size must be divisible by "
|
||||
f"{MACHETE_PREPACKED_BLOCK_SHAPE[0]}",
|
||||
)
|
||||
if out_featrues % MACHETE_PREPACKED_BLOCK_SHAPE[1] != 0:
|
||||
if out_features % MACHETE_PREPACKED_BLOCK_SHAPE[1] != 0:
|
||||
return (
|
||||
False,
|
||||
"Output features size must be divisible by "
|
||||
|
||||
Reference in New Issue
Block a user