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 13, 03:15 UTC
[Benchmark] Fix client seed synchronization in multi-turn benchmark (#28512)
Signed-off-by: ai-jz <aijz.xplr@gmail.com>
ai-jz committed
Nov 15, 2025 at 23:04 UTC
d231876ce31d8738a6e13a13591ae7d90d8b93f7
1 file changed
+6
-2
benchmarks/multi_turn/benchmark_serving_multi_turn.py
+6
-2
@@ -561,8 +561,11 @@ async def client_main(
561
f"{Color.CYAN}Started client {client_id}: max_num_requests={args.max_num_requests}, max_active_conversations={args.max_active_conversations}{Color.RESET}" # noqa: E501
562
)
563
564
- random.seed(args.seed)
565
- np.random.seed(args.seed)
564
+ # Set unique seed per client (each client runs in its own process)
565
+ # Add 1 to ensure no client uses the same seed as the main process
566
+ client_seed = args.seed + client_id + 1
567
+ random.seed(client_seed)
568
+ np.random.seed(client_seed)
569
570
# Active conversations
571
active_convs: ConversationsMap = {}
@@ -1490,6 +1493,7 @@ async def main() -> None:
1493
f"Invalid --warmup-percentage={args.warmup_percentage}"
1494
) from None
1495
1496
+ # Set global seeds for main process
1497
random.seed(args.seed)
1498
np.random.seed(args.seed)
1499