xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-24 16:35:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix CompressedTensorsWNA16MoE with grouped scales (#13769)

Browse Source
This commit is contained in:
Michael Goin 2025-02-25 03:17:14 -05:00 committed by GitHub
parent 18e505930d
commit 4d251ad00e
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/compressed_tensors/compressed_tensors_moe.py
View File

@ -527,7 +527,8 @@ class CompressedTensorsWNA16MoEMethod(CompressedTensorsMoEMethod):
replace_tensor("w13_weight_scale", marlin_w13_scales)
marlin_w2_scales = marlin_moe_permute_scales(
layer.w2_weight_scale,
layer.w2_weight_scale.shape[1] * self.packed_factor,
layer.w2_weight_scale.shape[1] *
(self.group_size if self.group_size != -1 else self.packed_factor),
size_k2,
self.group_size,
self.num_bits,