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

[Tiny] Fix grammar in shared experts fusion log messages (#18043)

Browse Source
This commit is contained in:
Mohammad Miadh Angkad
2026-02-01 05:14:25 +08:00
committed by GitHub
parent 46095f0551
commit 9ac4dcada4
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/models/deepseek_v2.py
View File

@@ -2844,7 +2844,7 @@ class DeepseekV2ForCausalLM(nn.Module, DeepseekV2WeightLoaderMixin):
or self.config.n_routed_experts != 256
or self.config.n_shared_experts != 1
):
disable_reason = "Config not support fused shared expert(s)."
disable_reason = "Config does not support fused shared expert(s)."
elif (not _is_cuda or torch.cuda.get_device_capability("cuda") < (8, 0)) and (
not _is_hip or torch.cuda.get_device_capability("cuda") < (9, 4)
):
@@ -2859,7 +2859,7 @@ class DeepseekV2ForCausalLM(nn.Module, DeepseekV2WeightLoaderMixin):
elif disable_reason is None and (
get_moe_a2a_backend().is_deepep() or get_moe_a2a_backend().is_mori()
):
disable_reason = "Deepseek V3/R1 can not use shared experts fusion optimization under deepep expert parallelism."
disable_reason = "Deepseek V3/R1 cannot use shared experts fusion optimization under deepep expert parallelism."
elif self.quant_config and self.quant_config.get_name() == "w4afp8":
disable_reason = "Deepseek V3/R1 W4AFP8 model uses different quant method for routed experts and shared experts."

2
python/sglang/srt/models/glm4_moe_lite.py
View File

@@ -531,7 +531,7 @@ class Glm4MoeLiteForCausalLM(DeepseekV2ForCausalLM):
):
disable_reason = "Only GLM-4.5 or GLM-4.6 on NV-platform with capability >= 80 can use shared experts fusion optimization."
elif get_moe_expert_parallel_world_size() > 1:
disable_reason = "GLM-4.5 or GLM-4.6 can not use shared experts fusion optimization under expert parallelism."
disable_reason = "GLM-4.5 or GLM-4.6 cannot use shared experts fusion optimization under expert parallelism."
if disable_reason is not None:
get_global_server_args().disable_shared_experts_fusion = True