From 1f2b84d28d639ba251f8f4c7b6ce94dea78e6605 Mon Sep 17 00:00:00 2001 From: YAMY <74099316+YAMY1234@users.noreply.github.com> Date: Mon, 1 Dec 2025 13:18:18 -0800 Subject: [PATCH] Fix NSA Bug in Centralize NSA Dispatch Logic (#14245) --- python/sglang/srt/layers/attention/nsa_backend.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/layers/attention/nsa_backend.py b/python/sglang/srt/layers/attention/nsa_backend.py index 5aa853010..172440336 100644 --- a/python/sglang/srt/layers/attention/nsa_backend.py +++ b/python/sglang/srt/layers/attention/nsa_backend.py @@ -1442,8 +1442,9 @@ class NativeSparseAttnBackend(AttentionBackend): # Requirements: H200/B200, short sequences, supported dtype, fits in chunk self.use_mha = ( - device_sm == 90 - or (device_sm >= 100 and device_sm < 110) # SM90/SM100f only + ( + device_sm == 90 or (device_sm >= 100 and device_sm < 110) + ) # SM90/SM100 only and max_kv_len <= self.nsa_index_topk # Short enough for MHA and forward_batch.token_to_kv_pool.dtype in [torch.bfloat16, torch.float8_e4m3fn]