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 9, 00:00 UTC
Fix offline test for Transformers v5 (#33682)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Feb 3, 2026 at 12:07 UTC
f6af34626d37f63ecb128e1f775ebcbbc1d0e5bf
1 file changed
+7
tests/entrypoints/offline_mode/test_offline_mode.py
+7
@@ -109,8 +109,15 @@ def _re_import_modules():
109
if k.startswith("transformers") and not k.startswith("transformers_modules")
110
]
111
112
+ # These modules are aliased in Transformers v5 and so cannot be reloaded directly
113
+ aliased_modules = ["tokenization_utils", "tokenization_utils_fast"]
114
+
115
reload_exception = None
116
for module_name in hf_hub_module_names + transformers_module_names:
117
+ if any(module_name.endswith(f".{alias}") for alias in aliased_modules):
118
+ # Remove from sys.modules so they are re-aliased on next import
119
+ del sys.modules[module_name]
120
+ continue
121
try:
122
importlib.reload(sys.modules[module_name])
123
except Exception as e: