Personal/ranxia/change max new tokens (#199)

* set multimodal llm max_new_tokens

* set multimodal llm max_new_tokens

src/pai_rag/modules/llm/multi_modal_llm.py

@@ -10,6 +10,8 @@
 logger = logging.getLogger(__name__)
 
 DEFAULT_DASHSCOPE_API_BASE = "https://dashscope.aliyuncs.com/compatible-mode/v1"
+DEFAULT_DASHSCOPE_MAX_NEW_TOKENS = 1500
+DEFAULT_EAS_MAX_NEW_TOKENS = 1500
 
 
 class MultiModalLlmModule(ConfigurableModule):
@@ -34,6 +36,7 @@ def _create_new_instance(self, new_params: Dict[str, Any]):
                 model=model_name,
                 api_base=DEFAULT_DASHSCOPE_API_BASE,
                 api_key=os.environ.get("DASHSCOPE_API_KEY"),
+                max_new_tokens=DEFAULT_DASHSCOPE_MAX_NEW_TOKENS,
             )
         elif llm_config.source.lower() == "paieas" and llm_config.get("endpoint"):
             logger.info("Using PAI-EAS Multi-Modal-LLM.")
@@ -43,6 +46,7 @@ def _create_new_instance(self, new_params: Dict[str, Any]):
                 ),  # TODO: change model path
                 api_base=llm_config.endpoint,
                 api_key=llm_config.token,
+                max_new_tokens=DEFAULT_EAS_MAX_NEW_TOKENS,
             )
         else:
             logger.info("Don't use Multi-Modal-LLM.")