EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[ci] fix llama4 ci error (#5126)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-04-07 21:15:46 +08:00
committed by GitHub
parent 57f99608f4
commit db452760e5
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -285,6 +285,7 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
activation: str = "silu", activation: str = "silu",
inplace: bool = True, inplace: bool = True,
no_combine: bool = False, no_combine: bool = False,
apply_router_weight_on_input: bool = False,
) -> torch.Tensor: ) -> torch.Tensor:
from sglang.srt.layers.moe.fused_moe_triton import fused_experts from sglang.srt.layers.moe.fused_moe_triton import fused_experts
from sglang.srt.layers.moe.topk import select_experts from sglang.srt.layers.moe.topk import select_experts
@@ -314,6 +315,7 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
w2_scale=layer.w2_weight_scale, w2_scale=layer.w2_weight_scale,
a1_scale=layer.w13_input_scale, a1_scale=layer.w13_input_scale,
a2_scale=layer.w2_input_scale, a2_scale=layer.w2_input_scale,
apply_router_weight_on_input=apply_router_weight_on_input,
) )