app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread
import torch

# 1. Load the Model and Tokenizer
# We use the specific 'openai-community/gpt2' as requested
model_id = "openai-community/gpt2"
print(f"Loading {model_id}...")

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

# 2. Define the Chat Logic
def chat_response(message, history):
    """
    This function handles the logic for generating a response.
    It formats the input to look like a conversation to help GPT-2 understand context.
    """
    
    # Construct a prompt based on conversation history
    # GPT-2 is a raw completion model, so we format it as a script:
    # User: Hello
    # AI: Hi there
    conversation = ""
    for user_msg, ai_msg in history:
        conversation += f"User: {user_msg}\nAI: {ai_msg}\n"
    
    # Add the current message
    prompt = conversation + f"User: {message}\nAI:"

    # Tokenize input
    inputs = tokenizer(prompt, return_tensors="pt")
    
    # Setup the streamer (this creates the 'typing' effect)
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    
    # Generation arguments
    generation_kwargs = dict(
        inputs=inputs.input_ids,
        max_new_tokens=100,      # Limit response length
        temperature=0.7,         # Controls randomness (0.7 is a good balance)
        top_p=0.9,               # Nucleus sampling for better quality
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id,
        streamer=streamer,
        no_repeat_ngram_size=2   # Prevents the model from repeating phrases
    )

    # Run generation in a separate thread so the GUI doesn't freeze
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    # Yield the text chunk by chunk to the Gradio interface
    generated_text = ""
    for new_text in streamer:
        generated_text += new_text
        yield generated_text

# 3. Create the GUI
# We use a custom theme to make it look professional
theme = gr.themes.Soft(
    primary_hue="purple",
    secondary_hue="cyan",
).set(
    body_background_fill="*neutral_50",
)

with gr.Blocks(theme=theme) as demo:
    gr.Markdown(
        """
        # 🤖 Classic GPT-2 Chat
        ### Powered by `openai-community/gpt2`
        Type a message below to start chatting! 
        *Note: GPT-2 is an older model (2019), so it may be whimsical or repetitive.*
        """
    )
    
    # The Chat Interface
    gr.ChatInterface(
        fn=chat_response,
        retry_btn=None,
        undo_btn=None,
        clear_btn="Clear Chat",
        examples=["Hello, how are you?", "Tell me a story about a dragon.", "What is the capital of France?"],
    )

# 4. Launch the App
if __name__ == "__main__":
    demo.launch()