app.py

import os
from datasets import load_dataset
import pandas as pd
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import gradio as gr

# Step 1: Load dataset and save to CSV (auto-training data)
def load_and_save_dataset():
    print("Loading dataset from Hugging Face...")
    dataset = load_dataset("HuggingFaceFW/fineweb", split="train")
    print("Saving dataset to data.csv...")
    dataset.to_csv("data.csv")
    print("Done! Data saved to data.csv.")
    return "Dataset loaded and saved to data.csv."

# Run on startup
load_and_save_dataset()

# Step 2: Load GPT-2 model for inference
model_name = "gpt2"  # or "distilgpt2" for faster inference
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Create generator pipeline
generator = pipeline('text-generation', model=model, tokenizer=tokenizer, device=-1)

# Function to generate responses
def generate_response(prompt):
    responses = generator(
        prompt,
        max_length=100,
        do_sample=True,
        temperature=0.7,
        top_k=50,
        top_p=0.95,
        num_return_sequences=1
    )
    return responses[0]['generated_text'].strip()

# Step 3: Build Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## GPT-2 Based AI Assistant with Dataset Loaded from Hugging Face")
    gr.Textbox(value="Loading dataset...", interactive=False, lines=2)
    fetch_button = gr.Button("Load Dataset and Save CSV")
    output_message = gr.Textbox()

    def fetch_and_confirm():
        msg = load_and_save_dataset()
        return msg

    fetch_button.click(fetch_and_confirm, outputs=output_message)

    gr.Markdown("### Ask the AI Assistant")
    prompt_input = gr.Textbox(label="Enter your prompt", placeholder="Say something...")
    response_output = gr.Textbox(label="Response", lines=10)

    def respond(prompt):
        return generate_response(prompt)

    gr.Button("Ask").click(respond, inputs=prompt_input, outputs=response_output)

demo.launch()