From 0f76976c3ccfe204565db9baf9d4a5ea4f37c37b Mon Sep 17 00:00:00 2001 From: Rain Jiang <96632942+rainj-me@users.noreply.github.com> Date: Fri, 7 Nov 2025 13:30:30 -0800 Subject: [PATCH] remove the fa4 page_size hardcode to 128 restriction on mla model arch (#12801) --- python/sglang/srt/server_args.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index 4367d729c..d025c1b73 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -1263,9 +1263,9 @@ class ServerArgs: raise ValueError( "FA4 backend is only supported for prefill. Please use `--prefill-attention-backend fa4` instead." ) - if self.prefill_attention_backend == "fa4": + if self.prefill_attention_backend == "fa4" and not self.use_mla_backend(): logger.warning( - f"FA4 backend only supports page size 128, changing page_size from {self.page_size} to 128." + f"FA4 backend only supports page size 128 for non-MLA model architectures, changing page_size from {self.page_size} to 128." ) self.page_size = 128