From d0cfec7ab919bfab261db419e17c768f08a24dc8 Mon Sep 17 00:00:00 2001
From: youkaichao <youkaichao@gmail.com>
Date: Tue, 25 Mar 2025 22:05:39 +0800
Subject: [PATCH] [bugfix] fix inductor cache on max_position_embeddings
 (#15436)

Signed-off-by: youkaichao <youkaichao@gmail.com>
---
 vllm/config.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/vllm/config.py b/vllm/config.py
index 7390ec59..87ede1e0 100644
--- a/vllm/config.py
+++ b/vllm/config.py
@@ -221,6 +221,9 @@ class ModelConfig:
         factors.append(self.trust_remote_code)
         factors.append(self.rope_scaling)
         factors.append(self.rope_theta)
+        # rope cos/sin cache depends on the max_position_embeddings
+        factors.append(
+            getattr(self.hf_config, "max_position_embeddings", "None"))
         return hashlib.sha256(str(factors).encode()).hexdigest()
 
     def __init__(