biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Add file lock for ModelScope download (#14060)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li
2025-03-01 14:10:28 +08:00
committed by GitHub
parent f64ffa8c25
commit 6a84164add
4 changed files with 40 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
vllm/transformers_utils/tokenizer.py
View File

@@ -150,16 +150,22 @@ def get_tokenizer(
# pylint: disable=C.
from modelscope.hub.snapshot_download import snapshot_download
# avoid circuit import
from vllm.model_executor.model_loader.weight_utils import get_lock
# Only set the tokenizer here, model will be downloaded on the workers.
if not os.path.exists(tokenizer_name):
tokenizer_path = snapshot_download(
model_id=tokenizer_name,
cache_dir=download_dir,
revision=revision,
local_files_only=huggingface_hub.constants.HF_HUB_OFFLINE,
# Ignore weights - we only need the tokenizer.
ignore_file_pattern=[".*.pt", ".*.safetensors", ".*.bin"])
tokenizer_name = tokenizer_path
# Use file lock to prevent multiple processes from
# downloading the same file at the same time.
with get_lock(tokenizer_name, download_dir):
tokenizer_path = snapshot_download(
model_id=tokenizer_name,
cache_dir=download_dir,
revision=revision,
local_files_only=huggingface_hub.constants.HF_HUB_OFFLINE,
# Ignore weights - we only need the tokenizer.
ignore_file_pattern=[".*.pt", ".*.safetensors", ".*.bin"])
tokenizer_name = tokenizer_path
if tokenizer_mode == "slow":
if kwargs.get("use_fast", False):