import gradio as gr
import spaces # Required for ZeroGPU hardware
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load your deployed model
model_id = "Zeesnal786/llama3-pakistani-fintech-3b"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    device_map="auto", 
    torch_dtype=torch.bfloat16
)

# Decorate the function so ZeroGPU knows to allocate hardware here
@spaces.GPU
def chat_function(question):
    messages = [{"role": "user", "content": question}]
    prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    
    outputs = model.generate(**inputs, max_new_tokens=256)
    answer = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
    return answer

# Create the Gradio interface
demo = gr.Interface(
    fn=chat_function,
    inputs=gr.Textbox(label="Your Banking Question", placeholder="e.g. How do I register on NayaPay?"),
    outputs=gr.Textbox(label="Model Answer"),
    title="Pakistani Fintech FAQ",
    description="Ask me any banking question regarding easypaisa, JS Bank, MCB, or NayaPay."
)

# Disable SSR to prevent the asyncio ValueError on startup
demo.launch(ssr_mode=False)