[MoE Refactor] Remove MoE DP chunking (#39107)

Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
bnellnm
2026-04-14 09:48:05 -04:00
committed by GitHub
co-authored by Robert Shaw
parent f7e62e3d66
commit e1e318af01
20 changed files with 76 additions and 389 deletions
+3
View File
@@ -32,6 +32,7 @@ from vllm.model_executor.layers.quantization.utils.flashinfer_utils import (
from vllm.model_executor.layers.quantization.utils.fp8_utils import input_to_float8
from vllm.model_executor.models.llama4 import Llama4MoE
from vllm.platforms import current_platform
from vllm.utils.math_utils import next_power_of_2
from vllm.utils.torch_utils import set_random_seed
try:
@@ -174,6 +175,7 @@ class TestData:
routing_method=layer.routing_method_type,
activation=activation,
device=w13_quantized.device,
max_num_tokens=next_power_of_2(m),
)
return TestData(
@@ -348,6 +350,7 @@ def test_flashinfer_cutlass_moe_fp8_no_graph(
in_dtype=torch.bfloat16,
is_act_and_mul=activation.is_gated,
routing_method=RoutingMethodType.TopK,
max_num_tokens=next_power_of_2(m),
)
kernel = mk.FusedMoEKernel(