Unverified Commit ad971af8 authored by zxfan-cpu's avatar zxfan-cpu Committed by GitHub
Browse files

[Bugfix] fix use-ep bug to enable ep by dp/tp size > 1 (#16161)

parent f2ebb6f5
...@@ -437,7 +437,7 @@ class FusedMoE(torch.nn.Module): ...@@ -437,7 +437,7 @@ class FusedMoE(torch.nn.Module):
# Use expert parallelism instead of tensor parallelism? # Use expert parallelism instead of tensor parallelism?
vllm_config = get_current_vllm_config() vllm_config = get_current_vllm_config()
use_ep = (vllm_config.parallel_config.enable_expert_parallel use_ep = (vllm_config.parallel_config.enable_expert_parallel
and self.tp_size > 1) and self.tp_size * self.dp_size > 1)
# For smuggling this layer into the fused moe custom op # For smuggling this layer into the fused moe custom op
self.use_direct_call = self.dp_size == 1 self.use_direct_call = self.dp_size == 1
......
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