app.py

import gradio as gr
import random
import time
from transformers import AutoModelForSequenceClassification
import os

with gr.Blocks() as demo:

    # Instalar accelerate
    os.system("pip install accelerate")

    # Instalar bitsandbytes
    os.system("pip install -i https://pypi.org/simple/ bitsandbytes")


    chatbot = gr.Chatbot()
    msg = gr.Textbox()
    clear = gr.ClearButton([msg, chatbot])

    model = AutoModelForSequenceClassification.from_pretrained("./modelo")

    tokenizer = AutoTokenizer.from_pretrained("./tokenizer")
    
    query_pipeline = transformers.pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        torch_dtype=torch.float16,
        device_map="auto", max_new_tokens=200)

    vectordb = Chroma.load(persist_directory="./chroma_db")

    def test_rag(pipeline, query):
      docs = vectordb.similarity_search_with_score(query)
      context = []
      for doc,score in docs:
        if(score<7):
          doc_details = doc.to_json()['kwargs']
          context.append( doc_details['page_content'])
      if(len(context)!=0):
        messages = [{"role": "user", "content": "Basándote en la siguiente información: " + "\n".join(context) + "\n Responde en castellano a la pregunta: " + query}]
        prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
        outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
        answer = outputs[0]["generated_text"]
        return answer[answer.rfind("[/INST]")+8:],docs
      else:
        return "No tengo información para responder a esta pregunta",docs

    
    def respond(message, chat_history):
        query = message
        answer, docs = test_rag(query_pipeline, query)
        chat_history.append((message, answer))
        time.sleep(2)
        return "", chat_history

    msg.submit(respond, [msg, chatbot], [msg, chatbot])

if __name__ == "__main__":
    demo.launch()