biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Refactor] Remove duplicate ceil_div (#20023)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2025-06-25 01:19:09 -04:00
committed by GitHub
parent 7108934142
commit 879f69bed3
7 changed files with 20 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/layers/quantization/utils/fp8_utils.py
View File

@@ -19,7 +19,7 @@ from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
CUTLASS_BLOCK_FP8_SUPPORTED)
from vllm.platforms import current_platform
from vllm.triton_utils import tl, triton
from vllm.utils import direct_register_custom_op
from vllm.utils import cdiv, direct_register_custom_op
logger = init_logger(__name__)
has_deep_gemm = importlib.util.find_spec("deep_gemm") is not None
@@ -158,12 +158,9 @@ def apply_w8a8_block_fp8_linear(
if current_platform.is_cuda():
if current_platform.has_device_capability(100):
def ceil_div(x: int, y: int) -> int:
return (x + y - 1) // y
use_cutlass = cutlass_block_fp8_supported and (
ceil_div(weight.shape[0], 128) == weight_scale.shape[0]
and ceil_div(weight.shape[1], 128) == weight_scale.shape[1])
cdiv(weight.shape[0], 128) == weight_scale.shape[0]
and cdiv(weight.shape[1], 128) == weight_scale.shape[1])
else:
# TODO: update this after switching to public sm90 block scale gemm
# as it also supports weight.shape % 128 != 0