[Feature] Avoid flashinfer autotune each time when vllm source change (#54794)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
@@ -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:
|
||||
|
||||
Reference in New Issue
Block a user