[Feature] Avoid flashinfer autotune each time when vllm source change (#54794)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-09-01 15:03:30 -04:00
committed by GitHub
parent d98bb2a879
commit 6c58595c2d
2 changed files with 9 additions and 6 deletions
+7 -3
View File
@@ -438,7 +438,7 @@ class VllmConfig:
remaining requests are aborted once the timeout is reached.
"""
def compute_hash(self) -> str:
def compute_hash(self, include_version: bool = True) -> str:
"""
WARNING: Whenever a new field is added to this config,
ensure that it is included in the factors list if
@@ -449,14 +449,18 @@ class VllmConfig:
graph from input ids/embeddings to the final hidden states,
excluding anything before input ids/embeddings and after
the final hidden states.
Args:
include_version: Include the vLLM version in the hash.
"""
factors: list[Any] = []
# summarize vllm config
vllm_factors: list[Any] = []
from vllm import __version__
if include_version:
from vllm import __version__
vllm_factors.append(__version__)
vllm_factors.append(__version__)
if self.model_config:
vllm_factors.append(self.model_config.compute_hash())
if (
@@ -10,15 +10,14 @@ from pathlib import Path
from typing import TYPE_CHECKING
import vllm.envs as envs
from vllm.compilation.caching import aot_compile_hash_factors
if TYPE_CHECKING:
from vllm.v1.worker.gpu_model_runner import GPUModelRunner
def flashinfer_autotune_cache_hash(runner: "GPUModelRunner") -> str:
factors = aot_compile_hash_factors(runner.vllm_config)
return hashlib.sha256(str(factors).encode()).hexdigest()
config_hash = runner.vllm_config.compute_hash(include_version=False)
return hashlib.sha256(config_hash.encode()).hexdigest()
def resolve_flashinfer_autotune_file(runner: "GPUModelRunner") -> Path: