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 17, 00:30 UTC
[Bugfix] Fix flag name for `max_seq_len_to_capture` (#4935)
Signed-off-by: kerthcet <kerthcet@gmail.com>
Kante Yin committed
May 22, 2024 at 00:30 UTC
14772eeb8e8ec76e5e70142d12a7332fcec28ccb
1 file changed
+2
-2
vllm/engine/arg_utils.py
+2
-2
@@ -341,9 +341,9 @@ class EngineArgs:
341
help='Maximum context length covered by CUDA '
342
'graphs. When a sequence has context length '
343
'larger than this, we fall back to eager mode. '
344
- '(DEPRECATED. Use --max-seq_len-to-capture instead'
344
+ '(DEPRECATED. Use --max-seq-len-to-capture instead'
345
')')
346
- parser.add_argument('--max-seq_len-to-capture',
346
+ parser.add_argument('--max-seq-len-to-capture',
347
type=int,
348
default=EngineArgs.max_seq_len_to_capture,
349
help='Maximum sequence length covered by CUDA '