chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[NPU]kimi k2 thinking bugfix (#19387)

Browse Source 
Co-authored-by: sglang-npu-bot <sglangnpu@163.com>
This commit is contained in:
KnightLTC
2026-02-27 15:36:38 +08:00
committed by GitHub
parent 4f0f6cd9d0
commit eef44ec916
2 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
python/sglang/srt/layers/moe/fused_moe_triton/layer.py
View File

@@ -292,6 +292,7 @@ class FusedMoE(torch.nn.Module):
else self.weight_loader_fused
),
with_bias=with_bias,
moe_intermediate_size=intermediate_size,
)
self.quant_method.create_moe_runner(self, self.moe_runner_config)

2
python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py
View File

@@ -503,7 +503,7 @@ class NPUCompressedTensorsW4A16Int4DynamicMoE(CompressedTensorsMoEScheme):
self.num_experts = num_experts
if (
extra_weight_attrs.get(
"intermediate_size_full", intermediate_size_per_partition
"moe_intermediate_size", intermediate_size_per_partition
)
// intermediate_size_per_partition
> 1