Merge pull request #7 from saltwick/fix/openai-serving-render-kwargs

fix: pass VLLM serving kwargs through OpenAIServingRender

Author: sharonyu-115

nemo_rl/models/generation/vllm/vllm_worker_async.py

@@ -462,14 +462,24 @@ async def preprocess_chat(
 
                 return res
 
+        http_server_kwargs = self.cfg["vllm_cfg"].get(
+            "http_server_serving_chat_kwargs", {}
+        )
         openai_serving_render = NeMoRLOpenAIServingRender(
             model_config=model_config,
             renderer=engine_client.renderer,
             io_processor=engine_client.io_processor,
             model_registry=openai_serving_models.registry,
             request_logger=None,
-            chat_template=None,
-            chat_template_content_format="auto",
+            chat_template=http_server_kwargs.get("chat_template", None),
+            chat_template_content_format=http_server_kwargs.get("chat_template_content_format", "auto"),
+            trust_request_chat_template=http_server_kwargs.get("trust_request_chat_template", False),
+            enable_auto_tools=http_server_kwargs.get("enable_auto_tools", False),
+            exclude_tools_when_tool_choice_none=http_server_kwargs.get("exclude_tools_when_tool_choice_none", False),
+            tool_parser=http_server_kwargs.get("tool_parser", None),
+            reasoning_parser=http_server_kwargs.get("reasoning_parser", None),
+            default_chat_template_kwargs=http_server_kwargs.get("default_chat_template_kwargs", None),
+            log_error_stack=http_server_kwargs.get("log_error_stack", False),
         )
 
         ########################################