app.py

import torch
import transformers
import gradio as gr
from ragatouille import RAGPretrainedModel
from huggingface_hub import InferenceClient

retrieve_results = 10 

generate_kwargs = dict(
    temperature = None,
    max_new_tokens = 512,
    top_p = None,
    do_sample = False,
    )

RAG = RAGPretrainedModel.from_index("colbert/indexes/arxiv_colbert")
try:
  gr.Info("Setting up retriever, please wait...")
  _ = RAG.search("what is Mistral?", k = 1)
  gr.Info("Retriever working successfully!")
except:
  gr.Warning("Retriever not working!")

mark_text = '# 🔍 Search Results\n'

def rag_cleaner(inp):
    rank = inp['rank']
    title = inp['document_metadata']['title']
    content = inp['content']
    return f"{rank}. <b> {title} </b> \n Abstract: {content}"

def get_prompt_text(question, context, formatted = True):
    if formatted:
      sys_instruction = f"Context:\n {context} \n Given the following scientific paper abstracts, take a deep breath and lets think step by step to answer what the question. Cite the titles of your sources when answering."
      message = f"Question: {question}"
      return f"<s>" + f"[INST] {sys_instruction} " +  f" {message} [/INST] </s> "
    return f"Context:\n {context} \n Given the following info, take a deep breath and lets think step by step to answer the question: {question}. Cite the titles of your sources when answering.\n\n"

def get_references(question, retriever, k = retrieve_results):
    rag_out = retriever.search(query=question, k=k)
    return rag_out

def get_rag(message):
        return get_references(message, RAG)

with gr.Blocks(theme = gr.themes.Soft()) as demo:
    header = gr.Markdown("# ArXiv RAG")
    with gr.Group():
      msg = gr.Textbox(label = 'Search')
      with gr.Accordion("Advanced Settings", open=False):
        with gr.Row(equal_height = True):
          llm_model = gr.Dropdown(choices = ['mistralai/Mixtral-8x7B-Instruct-v0.1','mistralai/Mistral-7B-Instruct-v0.2', 'None'], value = 'mistralai/Mistral-7B-Instruct-v0.2', label = 'LLM Model')
          llm_results = gr.Slider(minimum=4, maximum=10, value=5, step=1, interactive=True, label="Top n results to sent as context")

    output_text = gr.Textbox(show_label = True, container = True, label = 'LLM Answer', visible = True)
    input = gr.Textbox(show_label = False, visible = False)
    gr_md = gr.Markdown(mark_text)

    def update_with_rag_md(message, llm_results_use = 5):
        rag_out = get_rag(message)
        md_text_updated = mark_text
        for i in range(retrieve_results):
          rag_answer = rag_out[i]
          title = rag_answer['document_metadata']['title'].replace('\n','')

          score = round(rag_answer['score'], 2)
          paper_title = f'''### **{score}** | [{title}](https://arxiv.org/abs/{rag_answer['document_id']})\n'''
          paper_abs = rag_answer['content']
          authors = rag_answer['document_metadata']['authors'].replace('\n','')
          authors_formatted = f'*{authors}*' + ' \n\n'
        
          md_text_updated += paper_title + authors_formatted + paper_abs +  '\n---------------\n'+ '\n'
        prompt = get_prompt_text(message, '\n\n'.join(rag_cleaner(out) for out in rag_out[:llm_results_use]))
        return md_text_updated, prompt

    def ask_llm(prompt, llm_model_picked = 'mistralai/Mistral-7B-Instruct-v0.2'):
       if llm_model_picked == 'None':
          return gr.Textbox(visible = False)
       client = InferenceClient(llm_model_picked)
       output = client.text_generation(prompt, **generate_kwargs,  stream=False, details=False, return_full_text=False)
       output = output.lstrip(' \n') if output.lstrip().startswith('\n') else output
       return gr.Textbox(output, visible = True)

    msg.submit(update_with_rag_md, [msg, llm_results], [gr_md, input]).success(ask_llm, [input, llm_model], output_text)

demo.launch(debug = True)