app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "adarksky/biden-gpt2"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Prepare the input
    full_prompt = f"{system_message}\n\n"
    for user_msg, assistant_msg in history:
        full_prompt += f"Human: {user_msg}\nAssistant: {assistant_msg}\n"
    full_prompt += f"Human: {message}\nAssistant:"

    # Tokenize the input
    inputs = tokenizer(full_prompt, return_tensors="pt")
    input_ids = inputs["input_ids"]

    # Generate the response
    response = ""
    for _ in range(max_tokens):
        with torch.no_grad():
            outputs = model.generate(
                input_ids,
                max_new_tokens=1,
                do_sample=True,
                temperature=temperature,
                top_p=top_p,
            )
        
        new_token = outputs[0][-1]
        token_str = tokenizer.decode(new_token)
        
        if token_str == tokenizer.eos_token:
            break
        
        response += token_str
        input_ids = outputs
        
        yield response

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a american president", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

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