fixed bug: qa chain doesn't return any source doc

app_modules/llm_loader.py

@@ -90,13 +90,11 @@ class LLMLoader:
     streamer: any
     max_tokens_limit: int
 
-    def __init__(
-        self, llm_model_type, max_tokens_limit: int = 2048, lc_serve: bool = False
-    ):
+    def __init__(self, llm_model_type, lc_serve: bool = False):
         self.llm_model_type = llm_model_type
         self.llm = None
         self.streamer = None if lc_serve else TextIteratorStreamer("")
-        self.max_tokens_limit = max_tokens_limit
+        self.max_tokens_limit = 2048
         self.search_kwargs = {"k": 4}
 
     def _init_streamer(self, tokenizer, custom_handler):
@@ -196,6 +194,9 @@ class LLMLoader:
                 )
                 print(f"            HF auth token: {str(token)[-5:]}")
 
+                if "Llama-2" in MODEL_NAME_OR_PATH:
+                    self.max_tokens_limit = 4096
+
                 is_t5 = "t5" in MODEL_NAME_OR_PATH
                 temperature = (
                     0.01