A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 19:45 UTC
Fixes bench (#29058)
Signed-off-by: drisspg <drisspguessous@gmail.com>
Driss Guessous committed
Nov 20, 2025 at 13:21 UTC
3fd74189db13c9793325d9a36539d891873d1ae4
1 file changed
+2
-1
vllm/compilation/caching.py
+2
-1
@@ -12,6 +12,7 @@ from torch.utils import _pytree as pytree
12
13
import vllm.envs as envs
14
from vllm.config import VllmConfig, get_current_vllm_config
15
+from vllm.config.utils import hash_factors
16
from vllm.logger import init_logger
17
18
try:
@@ -138,7 +139,7 @@ def compilation_config_hash_factors(vllm_config: VllmConfig) -> list[str]:
139
factors = []
140
# 0. factors come from the env, for example, The values of
141
# VLLM_PP_LAYER_PARTITION will affect the computation graph.
141
- env_hash = envs.compute_hash()
142
+ env_hash = hash_factors(envs.compile_factors())
143
factors.append(env_hash)
144
145
# 1. factors come from the vllm_config (it mainly summarizes how the