From 27053aa5ed2d7e8f080720ec1ad96cbeb353bb59 Mon Sep 17 00:00:00 2001 From: Clint Date: Fri, 6 Mar 2026 02:54:29 -0600 Subject: [PATCH] Fix MLA decode path returning unwritten (padded) rows (#19902) --- python/sglang/srt/layers/attention/aiter_backend.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/python/sglang/srt/layers/attention/aiter_backend.py b/python/sglang/srt/layers/attention/aiter_backend.py index dd48ae19a..7a2fca522 100755 --- a/python/sglang/srt/layers/attention/aiter_backend.py +++ b/python/sglang/srt/layers/attention/aiter_backend.py @@ -1704,7 +1704,8 @@ class AiterAttnBackend(AttentionBackend): num_kv_splits=num_kv_splits, ) - return o[q_mask] + total_valid_q = int(qo_indptr[-1].item()) + return o[:total_valid_q] else: o = q.new_empty( (q.shape[0], layer.tp_q_head_num, layer.v_head_dim),