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 6, 00:45 UTC
Fix offline mode test for Transformers v5 (#38095)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Mar 25, 2026 at 11:39 UTC
ba2910f73a0341b46525a3441da2e7cb66698d58
1 file changed
+7
-2
tests/entrypoints/offline_mode/test_offline_mode.py
+7
-2
@@ -7,6 +7,7 @@ import importlib
7
import sys
8
9
import pytest
10
+import regex as re
11
import urllib3
12
13
from vllm import LLM
@@ -110,11 +111,15 @@ def _re_import_modules():
111
]
112
113
# These modules are aliased in Transformers v5 and so cannot be reloaded directly
113
- aliased_modules = ["tokenization_utils", "tokenization_utils_fast"]
114
+ aliased_module_patterns = [
115
+ r".+\.tokenization_utils$",
116
+ r".+\.tokenization_utils_fast$",
117
+ r".+\.models\..+\.image_processing_.+_fast$",
118
+ ]
119
120
reload_exception = None
121
for module_name in hf_hub_module_names + transformers_module_names:
117
- if any(module_name.endswith(f".{alias}") for alias in aliased_modules):
122
+ if any(re.match(pattern, module_name) for pattern in aliased_module_patterns):
123
# Remove from sys.modules so they are re-aliased on next import
124
del sys.modules[module_name]
125
continue