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),