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 14, 05:30 UTC
[Bugfix] Relax tokenizer regex for mixtral to include 'tokenizer.model' (#25964)
Signed-off-by: Bowen Bao <bowenbao@amd.com>
Bowen Bao committed
Oct 3, 2025 at 15:31 UTC
78b8015a4d0f07be1836d78dfe9bfa70ebbfd431
1 file changed
+11
-5
vllm/transformers_utils/tokenizers/mistral.py
+11
-5
@@ -122,15 +122,21 @@ def list_local_repo_files(repo_id: str, revision: Optional[str]) -> list[str]:
122
123
124
def find_tokenizer_file(files: list[str]):
125
+ # Accept both versioned (tokenizer.model.v3) and unversioned
126
+ # (tokenizer.model) forms, plus tekken.json and tokenizer.mm.model
127
+ # variants. Previous pattern only matched the versioned variants.
128
file_pattern = re.compile(
126
- r"^tokenizer\.model\.v.*$|^tekken\.json$|^tokenizer\.mm\.model\.v.*$")
129
+ r"^tokenizer\.model(\.v.*)?|tekken\.json|tokenizer\.mm\.model(\.v.*)?$"
130
+ )
131
132
matched_files = [file for file in files if file_pattern.match(file)]
133
if len(matched_files) > 1:
130
- raise OSError(
131
- f"Found {len(matched_files)} files matching the "
132
- f"pattern: `{file_pattern.pattern}`. Make sure only one Mistral "
133
- f"tokenizer is present in {files}.")
134
+ logger.warning(
135
+ "Multiple files matched pattern `%s`: %s. Using %s.",
136
+ file_pattern.pattern,
137
+ matched_files,
138
+ matched_files[0],
139
+ )
140
elif len(matched_files) == 0:
141
raise OSError(
142
f"Found {len(matched_files)} files matching the "