From 77e6dcbbfad8cfca6867663b164f038820f7a0be Mon Sep 17 00:00:00 2001
From: Shanshan Shen <467638484@qq.com>
Date: Wed, 4 Mar 2026 11:41:27 +0800
Subject: [PATCH] [PluggableLayer][MM] Add PluggableLayer for RelPosAttention
 (#33753)

Signed-off-by: shen-shanshan <467638484@qq.com>
---
 docs/design/custom_op.md                  | 2 ++
 vllm/model_executor/models/deepencoder.py | 7 ++++++-
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/docs/design/custom_op.md b/docs/design/custom_op.md
index 034736ec66..a62d033072 100644
--- a/docs/design/custom_op.md
+++ b/docs/design/custom_op.md
@@ -54,6 +54,8 @@ For example:
 --8<-- "vllm/model_executor/layers/attention/mm_encoder_attention.py:mm_encoder_attn"
 
 --8<-- "vllm/model_executor/layers/mla.py:multi_head_latent_attention"
+
+--8<-- "vllm/model_executor/models/deepencoder.py:rel_pos_attention"
 ```
 
 **2. Activation:**
diff --git a/vllm/model_executor/models/deepencoder.py b/vllm/model_executor/models/deepencoder.py
index f7ae4264f6..68c101460d 100644
--- a/vllm/model_executor/models/deepencoder.py
+++ b/vllm/model_executor/models/deepencoder.py
@@ -18,6 +18,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 from transformers import CLIPVisionConfig
 
+from vllm.model_executor.custom_op import PluggableLayer
 from vllm.model_executor.layers.attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.quantization import QuantizationConfig
@@ -263,9 +264,13 @@ class Block(nn.Module):
         return x
 
 
-class RelPosAttention(nn.Module):
+# --8<-- [start:rel_pos_attention]
+@PluggableLayer.register("rel_pos_attention")
+class RelPosAttention(PluggableLayer):
     """Multi-head Attention block with relative position embeddings."""
 
+    # --8<-- [end:rel_pos_attention]
+
     def __init__(
         self,
         dim: int,