From 7bac9c8cdb1b4cd7b8a9fce7890b6f73660cd2ef Mon Sep 17 00:00:00 2001 From: Liangsheng Yin Date: Sat, 11 Jul 2026 02:29:49 -0700 Subject: [PATCH] [Spec] Enable draft extend cuda graph for DeepSeek-V4 attention backend (#30853) --- python/sglang/srt/speculative/eagle_worker_v2.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/python/sglang/srt/speculative/eagle_worker_v2.py b/python/sglang/srt/speculative/eagle_worker_v2.py index 80526c3f5..bc2d7f183 100644 --- a/python/sglang/srt/speculative/eagle_worker_v2.py +++ b/python/sglang/srt/speculative/eagle_worker_v2.py @@ -471,6 +471,11 @@ class EagleDraftWorker(EagleDraftWorkerBase): ) graph_supported_backend_types.append(DeepseekSparseAttnBackend) + from sglang.srt.layers.attention.deepseek_v4_backend import ( + DeepseekV4AttnBackend, + ) + + graph_supported_backend_types.append(DeepseekV4AttnBackend) graph_supported_backend = isinstance( self.draft_extend_attn_backend,