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

bailingMoE: Fix Key error of deepep_mode (#11465)

Browse Source 
Signed-off-by: Michael Qiu <qiudayu.qdy@antgroup.com>
Co-authored-by: Mike_Qiu <qiudayu.qdy@antgroup.com>
This commit is contained in:
Mike Qiu
2025-10-12 20:42:59 +08:00
committed by GitHub
parent ec1cd90ac9
commit 99a0704a36
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/models/bailing_moe.py
View File

@@ -54,7 +54,7 @@ from sglang.srt.layers.linear import (
RowParallelLinear, RowParallelLinear,
) )
from sglang.srt.layers.logits_processor import LogitsProcessor from sglang.srt.layers.logits_processor import LogitsProcessor
from sglang.srt.layers.moe import get_moe_a2a_backend from sglang.srt.layers.moe import get_deepep_mode, get_moe_a2a_backend
from sglang.srt.layers.moe.ep_moe.layer import get_moe_impl_class from sglang.srt.layers.moe.ep_moe.layer import get_moe_impl_class
from sglang.srt.layers.moe.fused_moe_triton.layer import FusedMoE from sglang.srt.layers.moe.fused_moe_triton.layer import FusedMoE
from sglang.srt.layers.moe.token_dispatcher import DeepEPDispatcher from sglang.srt.layers.moe.token_dispatcher import DeepEPDispatcher
@@ -293,7 +293,7 @@ class BailingMoESparseMoeBlock(nn.Module):
num_local_experts=config.num_experts // self.tp_size, num_local_experts=config.num_experts // self.tp_size,
hidden_size=config.hidden_size, hidden_size=config.hidden_size,
params_dtype=config.torch_dtype, params_dtype=config.torch_dtype,
deepep_mode=DeepEPMode[global_server_args_dict["deepep_mode"]], deepep_mode=get_deepep_mode(),
async_finish=True, # TODO async_finish=True, # TODO
return_recv_hook=True, return_recv_hook=True,
) )