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

[Kernels] Clean up FusedMoeMethodBase and modular kernel setup. Remove extra arguments from modular kernel methods. (#22035)

Browse Source 
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
bnellnm
2025-08-15 14:46:00 -04:00
committed by GitHub
parent 48b01fd4d4
commit 8ad7285ea2
54 changed files with 2010 additions and 1293 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/model_executor/layers/fused_moe/fused_moe.py
View File

@@ -1394,9 +1394,9 @@ def fused_experts(hidden_states: torch.Tensor,
# E8M0 scale, which means we requantize the weight and input to the specific
# scale. Fallen back to cutlass or triton for some cases would cause
# accuracy issue.
should_use_deep_gemm = is_blackwell_deep_gemm_e8m0_used(
) or _valid_deep_gemm(hidden_states, w1, w2)
if (allow_deep_gemm and use_fp8_w8a8 and should_use_deep_gemm):
if (allow_deep_gemm and use_fp8_w8a8
and (is_blackwell_deep_gemm_e8m0_used()
or _valid_deep_gemm(hidden_states, w1, w2))):
assert apply_router_weight_on_input is False
assert is_act_and_mul, (
"DeepGemm only supports is_act_and_mul=True for now.")
@@ -1905,7 +1905,6 @@ class TritonExperts(mk.FusedMoEPermuteExpertsUnpermute):
workspace2: torch.Tensor,
expert_tokens_meta: Optional[mk.ExpertTokensMetadata],
apply_router_weight_on_input: bool,
extra_expert_args: Optional[dict[str, Any]],
):
# Check constraints.
if self.use_int4_w4a16: