fixed fused alltoall execute all reduce (#5109)
### What this PR does / why we need it?
fixed fused alltoall execute all reduce, when moe_comm_type is
MoECommType.FUSED_ALLTOALL
if moe_comm_type in {MoECommType.ALLTOALL, MoECommType.MC2,
MoECommType.FUSED_ALLTOALL} \
and not shared_expert_dp_enabled():
shared_out = tensor_model_parallel_all_reduce(shared_out)
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: AlvisGong <gwly0401@163.com>
Co-authored-by: Jade Zheng <zheng.shoujian@outlook.com>
This commit is contained in:
1 parent
78602eab4f
commit
ef8157a5f2
1 file changed
+1
-1
@@ -533,7 +533,7 @@ class AscendSharedFusedMoE(SharedFusedMoE, AscendFusedMoE):
|
||||
# NOTE: This is exactly the opposite of `maybe_all_reduce_tensor_model_parallel`
|
||||
forward_context = get_forward_context()
|
||||
moe_comm_type = forward_context.moe_comm_type
|
||||
if moe_comm_type in {MoECommType.ALLTOALL, MoECommType.MC2} \
|
||||
if moe_comm_type in {MoECommType.ALLTOALL, MoECommType.MC2, MoECommType.FUSED_ALLTOALL} \
|
||||
and not shared_expert_dp_enabled():
|
||||
shared_out = tensor_model_parallel_all_reduce(shared_out)
|
||||
else:
|
||||
|
||||
Reference in new issue
Block a user