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, 22:35 UTC
[Bugfix]: fix metadata file copy in test_sharded_state_loader (#21830)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Jul 31, 2025 at 14:22 UTC
3e36fcbee642f41278a4881c9e2bfbbd7c28e607
1 file changed
+5
-3
tests/test_sharded_state_loader.py
+5
-3
@@ -1,6 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
+import fnmatch
5
import multiprocessing as mp
6
import os
7
import shutil
@@ -64,9 +65,10 @@ def _run_writer(input_dir, output_dir, weights_patterns, **kwargs):
65
# Copy metadata files to output directory
66
for file in os.listdir(input_dir):
67
if os.path.isdir(os.path.join(input_dir, file)):
67
- continue
68
- if not any(file.endswith(ext) for ext in weights_patterns):
69
- shutil.copy(f"{input_dir}/{file}", output_dir)
68
+ shutil.copytree(os.path.join(input_dir, file),
69
+ os.path.join(output_dir, file))
70
+ elif not any(fnmatch.fnmatch(file, ext) for ext in weights_patterns):
71
+ shutil.copy(os.path.join(input_dir, file), output_dir)
72
73
74
def _run_generate(input_dir, queue: mp.Queue, **kwargs):