xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 19:14:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix][Performance] Restore flashinfer autotuning for all scenarios (#27904)

Browse Source
This commit is contained in:
Varun Sundar Rabindranath 2025-11-04 02:56:21 -05:00 committed by GitHub
parent 53f6e81dfd
commit 4022a9d279
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 14 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
tests/quantization/test_blackwell_moe.py
View File

@ -172,21 +172,9 @@ def test_gptoss_mxfp4mxfp8_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch
can_initialize("openai/gpt-oss-20b", hf_overrides=HF_OVERRIDE_TEXT) can_initialize("openai/gpt-oss-20b", hf_overrides=HF_OVERRIDE_TEXT)
def test_gptoss_dp2_mxfp4mxfp8_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): def test_gptoss_eager(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8", "1")
monkeypatch.setenv("VLLM_ALL2ALL_BACKEND", "deepep_high_throughput")
can_initialize( can_initialize(
"openai/gpt-oss-20b", "openai/gpt-oss-20b",
extra_args=["--data-parallel-size", "2", "--enable-expert-parallel"],
hf_overrides=HF_OVERRIDE_TEXT,
)
def test_gptoss_dp2_mxfp4bf16_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_BF16", "1")
monkeypatch.setenv("VLLM_ALL2ALL_BACKEND", "deepep_high_throughput")
can_initialize(
"openai/gpt-oss-20b",
extra_args=["--data-parallel-size", "2", "--enable-expert-parallel"],
hf_overrides=HF_OVERRIDE_TEXT, hf_overrides=HF_OVERRIDE_TEXT,
extra_args=["--enforce-eager"],
) )

9
vllm/model_executor/layers/fused_moe/trtllm_moe.py
View File

@ -127,10 +127,17 @@ class TrtLlmGenExperts(mk.FusedMoEPermuteExpertsUnpermute):
"routing_method_type": 1, "routing_method_type": 1,
"do_finalize": True, "do_finalize": True,
"output": output, "output": output,
"tune_max_num_tokens": self.max_capture_size, "tune_max_num_tokens": max(self.max_capture_size, 1),
} }
from flashinfer import trtllm_fp4_block_scale_routed_moe from flashinfer import trtllm_fp4_block_scale_routed_moe
from vllm.utils.flashinfer import autotune
with autotune(False):
# Enable autotune when,
# https://github.com/flashinfer-ai/flashinfer/issues/2023 is
# resolved.
trtllm_fp4_block_scale_routed_moe(**kwargs) trtllm_fp4_block_scale_routed_moe(**kwargs)
return output return output

4
vllm/model_executor/layers/quantization/mxfp4.py
View File

@ -1047,7 +1047,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
None, None,
1 if renormalize else 0, # routing_method_type, renormalize 1 if renormalize else 0, # routing_method_type, renormalize
True, # do finalize True, # do finalize
tune_max_num_tokens=self.max_capture_size, tune_max_num_tokens=max(self.max_capture_size, 1),
)[0] )[0]
return trtllm_gen_output return trtllm_gen_output
elif ( elif (
@ -1122,7 +1122,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
tp_rank=self.moe.tp_rank, tp_rank=self.moe.tp_rank,
ep_size=self.moe.ep_size, ep_size=self.moe.ep_size,
ep_rank=self.moe.ep_rank, ep_rank=self.moe.ep_rank,
tune_max_num_tokens=self.max_capture_size, tune_max_num_tokens=max(self.max_capture_size, 1),
**extra_kwargs, **extra_kwargs,
) )

27
vllm/model_executor/warmup/kernel_warmup.py
View File

@ -11,7 +11,6 @@ from typing import TYPE_CHECKING
import torch import torch
import vllm.envs as envs import vllm.envs as envs
from vllm.config import CUDAGraphMode, VllmConfig
from vllm.logger import init_logger from vllm.logger import init_logger
from vllm.model_executor.warmup.deep_gemm_warmup import deep_gemm_warmup from vllm.model_executor.warmup.deep_gemm_warmup import deep_gemm_warmup
from vllm.platforms import current_platform from vllm.platforms import current_platform
@ -25,26 +24,6 @@ if TYPE_CHECKING:
logger = init_logger(__name__) logger = init_logger(__name__)
def flashinfer_autotune_supported(vllm_config: VllmConfig) -> bool:
"""
Record known issues with vllm + flashinfer autotune here. Return True if
and only if flashinfer autotune will run through without issues.
"""
is_tp_or_dp = (vllm_config.parallel_config.data_parallel_size > 1) or (
vllm_config.parallel_config.tensor_parallel_size > 1
)
is_fi_mxfp4_backend = (
envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_BF16
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8_CUTLASS
) or (
current_platform.is_cuda() and current_platform.is_device_capability(100)
) # on >=sm100, default mxfp4 backend is flashinfer
is_eager = vllm_config.compilation_config.cudagraph_mode == CUDAGraphMode.NONE
return not (is_tp_or_dp and is_fi_mxfp4_backend and is_eager)
def kernel_warmup(worker: "Worker"): def kernel_warmup(worker: "Worker"):
# Deep GEMM warmup # Deep GEMM warmup
do_deep_gemm_warmup = ( do_deep_gemm_warmup = (
@ -58,11 +37,7 @@ def kernel_warmup(worker: "Worker"):
deep_gemm_warmup(model, max_tokens) deep_gemm_warmup(model, max_tokens)
# FlashInfer autotune for Hopper (SM 9.0) and Blackwell (SM 10.0) GPUs # FlashInfer autotune for Hopper (SM 9.0) and Blackwell (SM 10.0) GPUs
if ( if has_flashinfer() and current_platform.has_device_capability(90):
has_flashinfer()
and current_platform.has_device_capability(90)
and flashinfer_autotune_supported(worker.vllm_config)
):
flashinfer_autotune(worker.model_runner) flashinfer_autotune(worker.model_runner)
# FlashInfer attention warmup # FlashInfer attention warmup