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 16, 10:45 UTC
[Misc] Add random seed for prefix cache benchmark (#9081)
Andy Dai committed
Oct 4, 2024 at 14:58 UTC
0cc566ca8fd2d21a94f3a8e48bf5c5b60d42b59f
1 file changed
+3
-1
benchmarks/benchmark_prefix_caching.py
+3
-1
@@ -113,7 +113,7 @@ def repeat_and_sort_requests(requests: List[Tuple[str, int, int]],
113
def main(args):
114
tokenizer = get_tokenizer(args.model, trust_remote_code=True)
115
input_length_range = tuple(map(int, args.input_length_range.split(':')))
116
-
116
+ random.seed(args.seed)
117
if args.dataset_path is not None:
118
print(f"Start to sample {args.num_prompts} prompts"
119
"from {args.dataset_path}")
@@ -194,5 +194,7 @@ if __name__ == "__main__":
194
default='128:256',
195
help='Range of input lengths for sampling prompts,'
196
'specified as "min:max" (e.g., "128:256").')
197
+ parser.add_argument("--seed", type=int, default=0,
198
+ help='Random seed for reproducibility')
199
args = parser.parse_args()
200
main(args)