minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Fix nightly CI test for Kimi K2.5 INT4 + H200 (#28746)

Browse Source
This commit is contained in:
Brayden Zhong
2026-06-23 00:27:33 +00:00
committed by GitHub
parent d70726a31a
commit ba9d5aed98
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py
+5 -1
View File
@@ -130,7 +130,11 @@ class CompressedTensorsWNA16MoE(CompressedTensorsMoEScheme):
# In the case where we have actorder/g_idx, # In the case where we have actorder/g_idx,
# we do not partition the w2 scales # we do not partition the w2 scales
load_full_w2 = (self.actorder != "static") and self.group_size != -1 load_full_w2 = (
self.actorder is not None
and self.actorder != "static"
and self.group_size != -1
)
if load_full_w2: if load_full_w2:
w2_scales_size = intermediate_size_per_partition * layer.moe_tp_size w2_scales_size = intermediate_size_per_partition * layer.moe_tp_size