From 5d087891c93a6b66f0fd48b82fcf0a479d3e6ca5 Mon Sep 17 00:00:00 2001 From: kk <43161300+kkHuang-amd@users.noreply.github.com> Date: Thu, 18 Dec 2025 03:49:18 +0800 Subject: [PATCH] Fix dp run error with fp8-kv enable in high concurrency test (#15241) Co-authored-by: wunhuang --- python/sglang/srt/models/deepseek_v2.py | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-) diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py index 082feacf1..149597cf9 100644 --- a/python/sglang/srt/models/deepseek_v2.py +++ b/python/sglang/srt/models/deepseek_v2.py @@ -411,13 +411,7 @@ def handle_attention_trtllm_mla(attn, forward_batch): def handle_attention_aiter(attn, forward_batch): if forward_batch.forward_mode.is_extend_without_speculative(): - if is_dp_attention_enabled(): - if sum(forward_batch.extend_prefix_lens_cpu) == 0: - return AttnForwardMethod.MHA - else: - return AttnForwardMethod.MLA - else: - return AttnForwardMethod.MHA + return AttnForwardMethod.MHA else: return AttnForwardMethod.MLA