From 42ea9d01c94c7e59ced53789767450742c028384 Mon Sep 17 00:00:00 2001 From: ispobock Date: Sat, 14 Dec 2024 14:59:32 +0000 Subject: [PATCH] remove cuda graph bs adjust --- python/sglang/srt/server_args.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index 902f24ebb7f..943233a4f23 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -221,12 +221,10 @@ def __post_init__(self): if self.enable_dp_attention: self.dp_size = self.tp_size self.chunked_prefill_size = self.chunked_prefill_size // 2 - self.cuda_graph_max_bs = min(self.cuda_graph_max_bs, 96) self.schedule_conservativeness = self.schedule_conservativeness * 0.3 self.disable_overlap_schedule = True logger.warning( f"DP attention is enabled. The chunked prefill size is adjusted to {self.chunked_prefill_size} to avoid MoE kernel issues. " - f"The CUDA graph max batch size is adjusted to {self.cuda_graph_max_bs}. " f"The schedule conservativeness is adjusted to {self.schedule_conservativeness}. " "Data parallel size is adjusted to be the same as tensor parallel size. " "Overlap scheduler is disabled."