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

[WideEP] Remove pplx all2all backend (#33724)

Browse Source 
Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Tyler Michael Smith
2026-02-26 17:30:10 -05:00
committed by GitHub
parent 0f2f24c8b2
commit eb19955c37
39 changed files with 107 additions and 2069 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/layers/fused_moe/modular_kernel.py
View File

@@ -1172,9 +1172,9 @@ class FusedMoEModularKernel(torch.nn.Module):
# This happens when none of the tokens from the all2all reach this
# EP rank. Also, note that this is only relevant for CUDAGraph
# incompatible all2all kernels like the DeepEP high-throughput
# kernels. CUDAGraph compatible all2all kernels like the pplx
# kernels and the DeepEP low-latency kernels are always batched
# and can never run into the tensor.numel() == 0 case.
# kernels. CUDAGraph compatible all2all kernels like the DeepEP
# low-latency kernels are always batched and can never run into
# the tensor.numel() == 0 case.
if M_full == 0:
assert num_chunks == 0
workspace13 = None