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

Fix Mistral Large 3 nightly test (#25407)

Browse Source 
Co-authored-by: b8zhong <b8zhong@users.noreply.github.com>
This commit is contained in:
Brayden Zhong
2026-05-15 17:28:35 -07:00
committed by GitHub
co-authored by b8zhong
parent 5ba69f50fb
commit d523ae127f
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4_moe.py
+2 -2
View File
@@ -311,10 +311,10 @@ class CompressedTensorsW4A4Nvfp4MoE(CompressedTensorsMoEScheme):
router_logits = topk_output.router_logits router_logits = topk_output.router_logits
topk_config = topk_output.topk_config topk_config = topk_output.topk_config
# Quantize input hidden states using fp4_quantize # global_scale must be shape [1] (strict in cute-dsl backend).
hs_fp4_bytes, hs_sf_bytes = fp4_quantize( hs_fp4_bytes, hs_sf_bytes = fp4_quantize(
x, x,
layer.w13_input_scale_quant, layer.w13_input_scale_quant[:1],
self.group_size, # sf_vec_size self.group_size, # sf_vec_size
False, # use_ue8m0 False, # use_ue8m0
False, # is_sf_swizzled_layout False, # is_sf_swizzled_layout