A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 01:00 UTC
[Bug] DeepGemm: Fix TypeError: per_block_cast_to_fp8() missing 1 required positional argument: 'use_ue8m0' for SM100 (#21187)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Jul 19, 2025 at 02:25 UTC
37bd8d6e4c6e37e11ac69cc8844c57ab45dcee3c
1 file changed
+1
-1
vllm/utils/deep_gemm.py
+1
-1
@@ -99,7 +99,7 @@ def fp8_m_grouped_gemm_nt_masked(*args, **kwargs):
99
100
def per_block_cast_to_fp8(x, *args, **kwargs):
101
if _per_block_cast_impl is not None and is_blackwell_deep_gemm_used():
102
- return _per_block_cast_impl(x)
102
+ return _per_block_cast_impl(x, use_ue8m0=True)
103
# TODO: refactor the `per_block_cast_to_fp8` from tests to vllm utils
104
from tests.kernels.quant_utils import per_block_cast_to_fp8 as _pbcf
105
return _pbcf(x, *args, **kwargs)