import gradio as gr import os from langchain_community.document_loaders import JSONLoader from langchain_community.vectorstores import Qdrant from langchain_community.embeddings import HuggingFaceEmbeddings, HuggingFaceBgeEmbeddings from sentence_transformers.cross_encoder import CrossEncoder from groq import Groq client = Groq( api_key=os.environ.get("GROQ_API"), ) # loading data json_path = "format_food.json" json_path = "llama70b_food_dump.json" def metadata_func(record: dict, metadata: dict) -> dict: metadata["title"] = record.get("title") metadata["cuisine"] = record.get("cuisine") metadata["time"] = record.get("time") metadata["instructions"] = record.get("instructions") return metadata def reranking_results(query, top_k_results, rerank_model): # Load the model, here we use our base sized model top_results_formatted = [f"{item.metadata['title']}, {item.page_content}" for item in top_k_results] reranked_results = rerank_model.rank(query, top_results_formatted, return_documents=True) return reranked_results loader = JSONLoader( file_path=json_path, jq_schema='.dishes[].dish', text_content=False, content_key='doc', metadata_func=metadata_func ) data = loader.load() # Models # model_name = "Snowflake/snowflake-arctic-embed-xs" # rerank_model = CrossEncoder("mixedbread-ai/mxbai-rerank-xsmall-v1") # Embedding # model_kwargs = {"device": "cpu"} # encode_kwargs = {"normalize_embeddings": True} # hf_embedding = HuggingFaceEmbeddings( # model_name=model_name, # encode_kwargs=encode_kwargs, # model_kwargs=model_kwargs, # show_progress=True # ) model_name = "BAAI/bge-small-en" model_kwargs = {"device": "cpu"} encode_kwargs = {"normalize_embeddings": True} hf_embedding = HuggingFaceBgeEmbeddings( model_name=model_name, model_kwargs=model_kwargs, encode_kwargs=encode_kwargs ) qdrant = Qdrant.from_documents( data, hf_embedding, location=":memory:", # Local mode with in-memory storage only collection_name="my_documents", ) def format_to_markdown(response_list): response_list[0] = "- " + response_list[0] temp_string = "\n- ".join(response_list) return temp_string def run_query(query: str, groq: bool): print("Running Query") answer = qdrant.similarity_search(query=query, k=10) title_and_description = f"# Best Choice:\nA {answer[0].metadata['title']}: {answer[0].page_content}" instructions = format_to_markdown(answer[0].metadata['instructions']) recipe = f"# Standard Method\n## Cooking time:\n{answer[0].metadata['time']}\n\n## Recipe:\n{instructions}" print("Returning query") if groq: chat_completion = client.chat.completions.create( messages=[ { "role": "user", "content": f"please write a more detailed recipe for the following recipe:\n{recipe}\n\n please return it in the same format.", } ], model="Llama3-70b-8192", ) groq_update = "# Groq Update\n"+chat_completion.choices[0].message.content else: groq_update = "# Groq Update \nPlease select the tick box if you need more information." return title_and_description, recipe, groq_update with gr.Blocks() as demo: gr.Markdown("Start typing below and then click **Run** to see the output.") inp = gr.Textbox(placeholder="What sort of meal are you after?") groq_button = gr.Checkbox(value=False, label="Use Llama for a better recipe?") title_output = gr.Markdown(label="Title and description") instructions_output = gr.Markdown(label="Recipe") updated_recipe = gr.Markdown(label="Updated Recipe") btn = gr.Button("Run") btn.click(fn=run_query, inputs=[inp, groq_button], outputs=[title_output, instructions_output, updated_recipe]) demo.launch()