[bug] Fix lora reloading from disk every step when using AsyncLLM

vllm/entrypoints/openai/models/serving.py

@@ -196,6 +196,9 @@ async def load_lora_adapter(
                     ) from e
                 raise
 
+            # Reset load_inplace before caching so the flag does not leak to
+            # subsequent generate requests.
+            lora_request.load_inplace = False
             self.lora_requests[lora_name] = lora_request
             logger.info(
                 "Loaded new LoRA adapter: name '%s', path '%s'", lora_name, lora_path