import asyncio
import random
import gradio as gr
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

print("===== Application Startup =====")

# -----------------------
# Load model
# -----------------------
print("Loading model...")
model_name = "gpt2"  # you can swap this for a larger model if you have GPU
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
generator = pipeline("text-generation", model=model, tokenizer=tokenizer)
print("Model loaded successfully.")

# -----------------------
# Load dataset
# -----------------------
print("Fetching dataset...")
dataset = load_dataset("lvwerra/stack-exchange-paired", split="train[:200]")  
# limit to 200 for speed – you can increase if you want
print(f"Total prompts available: {len(dataset)}")

# Split dataset
initial_prompts = dataset[:20]   # first 20 for fast startup
remaining_prompts = dataset[20:] # remaining ~180

# Storage for loaded prompts
prompts = []
for item in initial_prompts:
    prompts.append(item["question"])

print(f"Loaded {len(prompts)} initial prompts for fast startup.")

# -----------------------
# Async loading of remaining prompts
# -----------------------
async def load_remaining_prompts():
    print("Background: Loading remaining prompts...")
    await asyncio.sleep(2)  # simulate delay
    for item in remaining_prompts:
        prompts.append(item["question"])
    print(f"Background: Finished loading. Total prompts now = {len(prompts)}")

# -----------------------
# Gradio interface
# -----------------------
def chat_with_model(user_input):
    """Respond to user with a random dataset prompt + model output."""
    if not prompts:
        return "Prompts not ready yet. Please wait..."
    prompt = random.choice(prompts)
    response = generator(f"{prompt}\n\nUser: {user_input}\nAI:", 
                         max_length=100, 
                         num_return_sequences=1,
                         do_sample=True)[0]["generated_text"]
    return response

demo = gr.Interface(
    fn=chat_with_model,
    inputs=gr.Textbox(lines=2, placeholder="Ask me something..."),
    outputs="text",
    title="Fast Prompt Loader Chatbot",
    description="Loads 20 prompts fast, then background loads 200+ prompts"
)

# -----------------------
# App runner
# -----------------------
if __name__ == "__main__":
    loop = asyncio.get_event_loop()
    loop.create_task(load_remaining_prompts())  # schedule async loading
    demo.launch(server_name="0.0.0.0", server_port=7860)