diff --git a/fastdeploy/cache_manager/cache_transfer_manager.py b/fastdeploy/cache_manager/cache_transfer_manager.py
index c9f062201..0aeed80af 100644
--- a/fastdeploy/cache_manager/cache_transfer_manager.py
+++ b/fastdeploy/cache_manager/cache_transfer_manager.py
@@ -382,4 +382,5 @@ if __name__ == "__main__":
 
     args = parse_args()
     logger = get_logger("cache_transfer_manager", "cache_transfer_manager.log")
+    paddle.set_device(f"gpu:{args.device_id}")
     main()
diff --git a/fastdeploy/engine/config.py b/fastdeploy/engine/config.py
index 035cea96c..ef7d11bd8 100644
--- a/fastdeploy/engine/config.py
+++ b/fastdeploy/engine/config.py
@@ -293,10 +293,11 @@ class Config:
         )
 
         if not self.cache_config.enable_chunked_prefill:
-            assert self.max_num_batched_tokens >= self.max_model_len, (
-                f"max_num_batched_tokens: {self.max_num_batched_tokens} "
-                f"should be larger than or equal to max_model_len: {self.max_model_len}"
-            )
+            if not int(os.getenv("FD_ENABLE_INTERNAL_ADAPTER", "0")):
+                assert self.max_num_batched_tokens >= self.max_model_len, (
+                    f"max_num_batched_tokens: {self.max_num_batched_tokens} "
+                    f"should be larger than or equal to max_model_len: {self.max_model_len}"
+                )
         else:
             assert self.max_num_batched_tokens >= self.cache_config.block_size, (
                 f"max_num_batched_tokens: {self.max_num_batched_tokens} "