From 81db702ed28d9a6edbd59fbd0ec039e107d36bc0 Mon Sep 17 00:00:00 2001
From: Jiangyun Zhu <riverclouds.zhu@qq.com>
Date: Tue, 25 Nov 2025 12:25:20 +0800
Subject: [PATCH] [Attention] add `_cudagraph_support` for linear attention
 (#28934)

Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
---
 vllm/v1/attention/backends/linear_attn.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/vllm/v1/attention/backends/linear_attn.py b/vllm/v1/attention/backends/linear_attn.py
index 1900c50849eca..004baa2d09cde 100644
--- a/vllm/v1/attention/backends/linear_attn.py
+++ b/vllm/v1/attention/backends/linear_attn.py
@@ -7,6 +7,7 @@ import torch
 from vllm.attention.backends.abstract import AttentionBackend
 from vllm.config import VllmConfig
 from vllm.v1.attention.backends.utils import (
+    AttentionCGSupport,
     AttentionMetadataBuilder,
     CommonAttentionMetadata,
     split_decodes_and_prefills,
@@ -35,6 +36,8 @@ class LinearAttentionMetadata:
 class LinearAttentionMetadataBuilder(AttentionMetadataBuilder[LinearAttentionMetadata]):
     reorder_batch_threshold: int = 1
 
+    _cudagraph_support = AttentionCGSupport.UNIFORM_SINGLE_TOKEN_DECODE
+
     def __init__(
         self,
         kv_cache_spec: AttentionSpec,