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, 23:45 UTC
[Misc] add default value for file pattern arg (#21659)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Jul 27, 2025 at 13:14 UTC
1cbf951ba272c230823b947631065b826409fa62
1 file changed
+5
-1
examples/offline_inference/save_sharded_state.py
+5
-1
@@ -29,6 +29,7 @@ import shutil
29
from pathlib import Path
30
31
from vllm import LLM, EngineArgs
32
+from vllm.model_executor.model_loader import ShardedStateLoader
33
from vllm.utils import FlexibleArgumentParser
34
35
@@ -39,7 +40,10 @@ def parse_args():
40
"--output", "-o", required=True, type=str, help="path to output checkpoint"
41
)
42
parser.add_argument(
42
- "--file-pattern", type=str, help="string pattern of saved filenames"
43
+ "--file-pattern",
44
+ type=str,
45
+ default=ShardedStateLoader.DEFAULT_PATTERN,
46
+ help="string pattern of saved filenames",
47
)
48
parser.add_argument(
49
"--max-file-size",