xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-06 17:44:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Skip the Q component for QKVParallelLinear in the case of QKVCrossParallelLinear since its width is 0 (#22369)

Browse Source 
Signed-off-by: sstamenk <sstamenk@amd.com>
This commit is contained in:
sstamenk 2025-08-15 19:17:31 +02:00 committed by GitHub
parent 1c859a1387
commit 6b04039a72
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/quantization/utils/w8a8_utils.py
View File

@ -121,6 +121,9 @@ def requantize_with_max_scale(
if unfused_module_in_checkpoint:
start = 0
for idx, logical_width in enumerate(logical_widths):
# Skip any component with zero width.
if logical_width == 0:
continue
end = start + logical_width
weight_dq = per_tensor_dequantize(weight[start:end, :],
weight_scale[idx])