biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Kernel] Enable fp8 support for pplx and BatchedTritonExperts. (#18864)

Browse Source 
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2025-07-03 17:55:40 -04:00
committed by GitHub
parent 2f2fcb31b8
commit 78fe77534b
25 changed files with 1277 additions and 663 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/models/qwen3_moe.py
View File

@@ -135,7 +135,7 @@ class Qwen3MoeSparseMoeBlock(nn.Module):
router_logits, _ = self.gate(hidden_states)
final_hidden_states = self.experts(hidden_states=hidden_states,
router_logits=router_logits)
final_hidden_states = final_hidden_states
if self.tp_size > 1:
final_hidden_states = self.experts.maybe_all_reduce_tensor_model_parallel( # noqa E501
final_hidden_states)