EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

fix nz for quantization (#4943)

Browse Source 
quantization ops rely on NZ by force, we should remove the nz check for it.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-12-12 14:54:41 +08:00
committed by GitHub
parent 5932abc446
commit 01a13a9b77
4 changed files with 11 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm_ascend/quantization/w8a8_dynamic.py
View File

@@ -270,9 +270,8 @@ class AscendW8A8DynamicFusedMoEMethod:
1, 2).contiguous()
layer.w2_weight.data = layer.w2_weight.data.transpose(
1, 2).contiguous()
if is_enable_nz():
torch_npu.npu_format_cast_(layer.w13_weight, ACL_FORMAT_FRACTAL_NZ)
torch_npu.npu_format_cast_(layer.w2_weight, ACL_FORMAT_FRACTAL_NZ)
torch_npu.npu_format_cast_(layer.w13_weight, ACL_FORMAT_FRACTAL_NZ)
torch_npu.npu_format_cast_(layer.w2_weight, ACL_FORMAT_FRACTAL_NZ)
layer.w13_weight_scale.data = layer.w13_weight_scale.data.view(
layer.w13_weight_scale.data.shape[0], -1)
layer.w13_weight_scale_fp32 = layer.w13_weight_scale.data.to(