Update with h2oGPT hash 6a58e8f0a87c4fc67f2b2bec736cf34f8c1bbc39

gen.py

@@ -403,13 +403,13 @@ def main(
 
     # auto-set langchain_mode
     if have_langchain and langchain_mode is None:
+        # start in chat mode, in case just want to chat and don't want to get "No documents to query" by default.
+        langchain_mode = LangChainMode.CHAT_LLM.value
         if allow_upload_to_user_data and not is_public and user_path:
-            langchain_mode = 'UserData'
-            print("Auto set langchain_mode=%s" % langchain_mode, flush=True)
+            print("Auto set langchain_mode=%s.  Could use UserData instead." % langchain_mode, flush=True)
         elif allow_upload_to_my_data:
-            langchain_mode = 'MyData'
-            print("Auto set langchain_mode=%s."
-                  "  To use UserData to pull files from disk,"
+            print("Auto set langchain_mode=%s.  Could use MyData instead."
+                  "  To allow UserData to pull files from disk,"
                   " set user_path and ensure allow_upload_to_user_data=True" % langchain_mode, flush=True)
         else:
             raise RuntimeError("Please pass --langchain_mode=<chosen mode> out of %s" % langchain_modes)
@@ -1555,7 +1555,8 @@ def evaluate(
             where_from = "openai_client"
 
             openai.api_key = os.getenv("OPENAI_API_KEY")
-            stop_sequences = list(set(prompter.terminate_response + [prompter.PreResponse]))
+            terminate_response = prompter.terminate_response or []
+            stop_sequences = list(set(terminate_response + [prompter.PreResponse]))
             stop_sequences = [x for x in stop_sequences if x]
             # OpenAI will complain if ask for too many new tokens, takes it as min in some sense, wrongly so.
             max_new_tokens_openai = min(max_new_tokens, model_max_length - num_prompt_tokens)
@@ -1764,7 +1765,8 @@ def evaluate(
 
                 # prompt must include all human-bot like tokens, already added by prompt
                 # https://github.com/huggingface/text-generation-inference/tree/main/clients/python#types
-                stop_sequences = list(set(prompter.terminate_response + [prompter.PreResponse]))
+                terminate_response = prompter.terminate_response or []
+                stop_sequences = list(set(terminate_response + [prompter.PreResponse]))
                 stop_sequences = [x for x in stop_sequences if x]
                 gen_server_kwargs = dict(do_sample=do_sample,
                                          max_new_tokens=max_new_tokens,

gpt_langchain.py

@@ -643,7 +643,8 @@ def get_llm(use_openai_model=False,
 
         callbacks = [StreamingGradioCallbackHandler()]
         assert prompter is not None
-        stop_sequences = list(set(prompter.terminate_response + [prompter.PreResponse]))
+        terminate_response = prompter.terminate_response or []
+        stop_sequences = list(set(terminate_response + [prompter.PreResponse]))
         stop_sequences = [x for x in stop_sequences if x]
 
         if gr_client: