From 49796a4657fdcc5b1325af58196e4010ee92f407 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Niklas=20M=C3=BCller?= Date: Sun, 23 Jun 2024 19:27:11 +0200 Subject: [PATCH] remove lite_llm; distinguish if first message or whole conversation in chat querifier; Add info to README --- .env | 3 +-- rag-chat-backend/src/endpoints/llm.py | 13 ++++++++++--- 2 files changed, 11 insertions(+), 5 deletions(-) diff --git a/.env b/.env index 7fe3903..31d7180 100644 --- a/.env +++ b/.env @@ -42,6 +42,5 @@ LLM_API_KEY=placeholder # valid Language options - "en" "de" "multi" LLM_LANGUAGE=de -# LLM Model to be used - "mistral", "phi3", "llama3", etc. +# LLM Model to be used - "mistral", "phi3", "llama3", etc. (Check available with :11434/api/tags) LLM_MODEL_NAME=mistral - diff --git a/rag-chat-backend/src/endpoints/llm.py b/rag-chat-backend/src/endpoints/llm.py index 6168d96..90e0a52 100644 --- a/rag-chat-backend/src/endpoints/llm.py +++ b/rag-chat-backend/src/endpoints/llm.py @@ -38,7 +38,6 @@ SEARCH_COMPONENT = IndexSearchComponent(os_client=OS_INTERFACE) PROMPTS = LLM.prompt_obj llm_component_name = "ollama-llm" selected_language = "german" -lite_llm = os.getenv("LITE_LLM") stream_llms = ["fu"] @@ -97,7 +96,15 @@ def chat(stream_response, messages, tags=None, languages: str = None, start_date raise HTTPException(status_code=400, detail="Invalid format for 'messages'") # provides you a query based on the whole chat - query = LLM.llm_chat_querifier(chat=messages) + if len(messages) <= 2: + if messages[0]["role"] == "user": + query = messages[0]["content"] + elif messages[1]["role"] == "user": + query = messages[1]["content"] + else: + query = LLM.llm_chat_querifier(chat=messages) + else: + query = LLM.llm_chat_querifier(chat=messages) logger.info("Generated query: '%s'", query) # creating a new dashboard logger at the start of the project @@ -137,7 +144,7 @@ def chat(stream_response, messages, tags=None, languages: str = None, start_date # check if chat is allowed for selected llm logger.info("Stream response: %s", stream_response) if stream_response == "true": - if lite_llm == "true" or llm_component_name in stream_llms: + if llm_component_name in stream_llms: answer = LLM.llm_chat_request( chat=messages, context=formatted_context,