xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-27 09:28:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bug] Fix B200 DeepGEMM E8M0 Accuracy Issue (#22399)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-08-06 20:03:53 -04:00 committed by GitHub
parent 46a13949d5
commit eec890c1c1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/quantization/utils/fp8_utils.py
View File

@ -799,7 +799,8 @@ def requant_weight_ue8m0_inplace(
s_exp = s_exp[:m_cur, :k_cur]
w_dq = w_q.to(torch.float32) * s_exp
# Re-quantise using power-of-two scaling (UE8M0).
w_requant, s_requant = per_block_cast_to_fp8(w_dq, [block_m, block_k])
w_requant, s_requant = per_block_cast_to_fp8(w_dq, [block_m, block_k],
use_ue8m0=True)
# Write back the results in-place.
w_q.copy_(w_requant)