Skip to content

GitLab

Unverified Commit 476c67d7 authored by Shangming Cai's avatar Shangming Cai Committed by GitHub
Browse files

Fix missing a2a backend init of GLM4.5 MoE Block (#11692) 


Signed-off-by: default avatarShangming Cai <csmthu@gmail.com>
parent 3289da5b
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 2 deletions
python/sglang/srt/models/glm4_moe.py
View file @ 476c67d7
...@@ -467,7 +467,7 @@ class Glm4MoeSparseMoeBlock(DeepseekV2MoE): ...@@ -467,7 +467,7 @@ class Glm4MoeSparseMoeBlock(DeepseekV2MoE):
self.top_k = config.num_experts_per_tok self.top_k = config.num_experts_per_tok
if get_moe_a2a_backend().is_deepep(): if get_moe_a2a_backend().is_deepep() or get_moe_a2a_backend().is_mooncake():
# TODO: we will support tp < ep in the future # TODO: we will support tp < ep in the future
self.ep_size = get_moe_expert_parallel_world_size() self.ep_size = get_moe_expert_parallel_world_size()
self.num_experts = ( self.num_experts = (
...@@ -496,7 +496,9 @@ class Glm4MoeSparseMoeBlock(DeepseekV2MoE): ...@@ -496,7 +496,9 @@ class Glm4MoeSparseMoeBlock(DeepseekV2MoE):
return_recv_hook=True, return_recv_hook=True,
) )
self._enable_deepep_moe = get_moe_a2a_backend().is_deepep() self._enable_a2a_moe = (
get_moe_a2a_backend().is_deepep() or get_moe_a2a_backend().is_mooncake()
)
def forward_normal_dual_stream( def forward_normal_dual_stream(
self, self,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment