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

[Core] Default to using per_token quantization for fp8 when cutlass is supported. (#8651)

Browse Source 
Signed-off-by: mgoin <michael@neuralmagic.com>
Co-authored-by: Michael Goin <mgoin@redhat.com>
Co-authored-by: mgoin <michael@neuralmagic.com>
This commit is contained in:
Elfie Guo
2025-01-15 20:31:27 -08:00
committed by GitHub
parent cd9d06fb8d
commit fa0050db08
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/quantization/fp8.py
View File

@@ -355,7 +355,8 @@ class Fp8LinearMethod(LinearMethodBase):
input_scale=layer.input_scale,
bias=bias,
cutlass_fp8_supported=self.cutlass_fp8_supported,
use_per_token_if_dynamic=False)
# Default to using per_token quantization if cutlass is supported
use_per_token_if_dynamic=self.cutlass_fp8_supported)
class Fp8MoEMethod(FusedMoEMethodBase):