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

[Kernel] FlashInfer: switch allreduce fusion to unified API (#33985)

Browse Source 
Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
This commit is contained in:
Mohammad Miadh Angkad
2026-02-09 23:43:24 +08:00
committed by GitHub
parent cb62e86f83
commit d4f123cc48
3 changed files with 80 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/compile/passes/distributed/test_fusion_all_reduce.py
View File

@@ -202,9 +202,10 @@ class TestAllReduceFusedAddRMSNormStaticQuantFP4Model(torch.nn.Module):
@pytest.mark.skipif(envs.VLLM_TARGET_DEVICE not in ["cuda"], reason="Only test on CUDA")
@pytest.mark.skipif(
not find_spec("flashinfer")
or not has_module_attribute("flashinfer.comm", "trtllm_allreduce_fusion"),
or not has_module_attribute("flashinfer.comm", "allreduce_fusion")
or not has_module_attribute("flashinfer.comm", "create_allreduce_fusion_workspace"),
reason="flashinfer is not found or flashinfer "
"is not compiled with trtllm_allreduce_fusion",
"is not compiled with allreduce_fusion",
)
def test_all_reduce_fusion_pass_replace(
test_model: torch.nn.Module,