Revert "[Bugfix][MoE] Tune FlashInfer experts to scheduler token limit" (#52989) (#53186)

Co-authored-by: vllm-ci-failure-analyzer <kevin@inferact.ai>
This commit is contained in:
vllm-agent
2026-08-21 12:39:28 -06:00
committed by GitHub
co-authored by vllm-ci-failure-analyzer
parent 7a2fdbaac4
commit ba53da60bb
6 changed files with 1 additions and 12 deletions
-2
View File
@@ -72,7 +72,6 @@ def test_flashinfer_swigluoai_params_are_forwarded(activation, monkeypatch):
moe_parallel_config=FusedMoEParallelConfig.make_no_parallel(),
in_dtype=torch.bfloat16,
routing_method=RoutingMethodType.TopK,
max_num_tokens=16_384,
)
quant_config = FusedMoEQuantConfig.make(
gemm1_alpha=1.702,
@@ -111,7 +110,6 @@ def test_flashinfer_swigluoai_params_are_forwarded(activation, monkeypatch):
assert experts._supports_activation(activation)
assert call_args["activation_type"] == ActivationType.Swiglu
assert call_args["tune_max_num_tokens"] == 16_384
for name, value in (
("swiglu_alpha", 1.702),
("swiglu_beta", 1.0),