xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-25 21:17:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

lint

Browse Source 
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
Bill Nell 2025-05-28 23:38:19 +00:00
parent 3ca8322b74
commit d86e3f0172
1 changed files with 4 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/model_executor/layers/fused_moe/fused_batched_moe.py
View File

@ -614,13 +614,10 @@ class BatchedExperts(mk.FusedMoEPermuteExpertsUnpermute):
else:
num = int(expert_num_tokens[expert].item())
tmp = _resize_cache(workspace2, (num, N))
if self.use_fp8_w8a8:
assert False # TBD
else:
input = hidden_states[expert, :num, :] @ w1[expert].transpose(
0, 1)
self.activation(activation, tmp, input)
out[expert, :num, :] = tmp @ w2[expert].transpose(0, 1)
assert not self.use_fp8_w8a8
input = hidden_states[expert, :num, :] @ w1[expert].transpose(0, 1)
self.activation(activation, tmp, input)
out[expert, :num, :] = tmp @ w2[expert].transpose(0, 1)
return out