From 10024a6282183d9ae4847829af920746241536cd Mon Sep 17 00:00:00 2001
From: xffxff <1247714429@qq.com>
Date: Wed, 22 Jan 2025 17:42:51 +0800
Subject: [PATCH] add some comments

Signed-off-by: xffxff <1247714429@qq.com>
---
 vllm/model_executor/models/aria.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/vllm/model_executor/models/aria.py b/vllm/model_executor/models/aria.py
index d5179df3de1fc..8c6873de13627 100644
--- a/vllm/model_executor/models/aria.py
+++ b/vllm/model_executor/models/aria.py
@@ -60,6 +60,9 @@ def __init__(
         prefix: str = "",
     ) -> None:
         super().__init__(config, quant_config, prefix)
+        # Unlike Idefics3VisionTransformer which uses LayerNorm after the
+        # final layer, Aria omits this normalization, so we replace it with an
+        # Identity layer
         self.post_layernorm = nn.Identity()
 
     def load_weights(self, weights: Iterable[Tuple[str,