app.py

import gradio as gr
import requests
from pdfminer.high_level import extract_text
from langchain_community.vectorstores import Chroma
from langchain_huggingface import HuggingFaceEmbeddings, ChatHuggingFace
from langchain_core.runnables import RunnablePassthrough, Runnable
from io import BytesIO
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser
from langchain_core.documents import Document
from langchain_core.prompts import ChatPromptTemplate
from langchain.text_splitter import CharacterTextSplitter
from huggingface_hub import InferenceClient
import time
from tenacity import retry, stop_after_attempt, wait_exponential, retry_if_exception_type
import logging
import os


# lo = "hf_JyAJApaXhIrONPFSIo"
# ve = "wbnJbrXViYurrsvP"

last_call_time = 0
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "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")

def extract_pdf_text(url: str) -> str:
    response = requests.get(url)
    pdf_file = BytesIO(response.content)
    text = extract_text(pdf_file)
    return text


pdf_url = "https://arxiv.org/pdf/2408.09869"
text = extract_pdf_text(pdf_url)
docs_list = [Document(page_content=text)]

text_splitter = CharacterTextSplitter.from_tiktoken_encoder(chunk_size=7500, chunk_overlap=100)
docs_splits = text_splitter.split_documents(docs_list)

embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
vectorstore = Chroma.from_documents(
    documents=docs_splits,
    collection_name="rag-chroma",
    embedding=embeddings,
)
retriever = vectorstore.as_retriever()

llm = ChatOpenAI(
    model="gpt-3.5-turbo",
    api_key=OPENAI_API_KEY,
    max_tokens=512,
    temperature=0.7,
    top_p=0.95
)

# After RAG chain
after_rag_template = """You are a {role}. Summarize the following content for yourself and speak in terms of first person.
Only include content relevant to that role like a resume summary.

Context:
{context}

Question: Give a one paragraph summary of the key skills a {role} can have from this document.
"""
after_rag_prompt = ChatPromptTemplate.from_template(after_rag_template)

def format_query(input_dict):
    return f"Give a one paragraph summary of the key skills a {input_dict['role']} can have from this document."

after_rag_chain = (
    {
        "context": format_query | retriever,
        "role": lambda x: x["role"],
    }
    | after_rag_prompt
    | llm
    | StrOutputParser()
)

def process_query(role, system_message, max_tokens, temperature, top_p):

    global last_call_time
    current_time = time.time()
    
    if current_time - last_call_time < 60:
        wait_time = int(60 - (current_time - last_call_time))
        return f"Rate limit exceeded. Please wait {wait_time} seconds before trying again."
    # llm.update_params(max_tokens, temperature, top_p)
    last_call_time = current_time
    llm.max_tokens = max_tokens
    llm.temperature = temperature
    llm.top_p = top_p

    # After RAG
    after_rag_result = after_rag_chain.invoke({"role": role})

    return f"**RAG Summary**\n{after_rag_result}"


with gr.Blocks() as demo:
    gr.Markdown("## Zephyr Chatbot Controls")

    role_dropdown = gr.Dropdown(choices=["SDE", "BA"], label="Select Role", value="SDE")

    system_message = gr.Textbox(value="You are a friendly chatbot.", label="System message")
    max_tokens = gr.Slider(1, 2048, value=512, label="Max tokens")
    temperature = gr.Slider(0.1, 4.0, value=0.7, label="Temperature", step=0.1)
    top_p = gr.Slider(0.1, 1.0, value=0.95, label="Top-p", step=0.05)

    output = gr.Textbox(label="Output", lines=20)

    submit_btn = gr.Button("Submit")
    clear_btn = gr.Button("Clear")

    submit_btn.click(
        fn=process_query,
        inputs=[role_dropdown, system_message, max_tokens, temperature, top_p],
        outputs=output
    )

    clear_btn.click(
        fn=lambda: ("", gr.Info("Chat cleared!")),
        outputs=[output]
    )

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