import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import gradio as gr
import spaces

huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
if not huggingface_token:
    raise ValueError("HUGGINGFACE_TOKEN environment variable is not set")

model_id = "tokyotech-llm/Llama-3.1-Swallow-8B-Instruct-v0.1"
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16

tokenizer = AutoTokenizer.from_pretrained(model_id, token=huggingface_token)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map=device,
    torch_dtype=dtype,
    token=huggingface_token
)

@spaces.GPU
def generate_text(prompt, system_message="あなたは誠実で優秀な日本人のアシスタントです。"):
    messages = [
        {"role": "user", "content": prompt},
    ]
    
    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(model.device)
    
    outputs = model.generate(inputs, max_new_tokens=1024, do_sample=True, temperature=0.7)
    generated_text = tokenizer.batch_decode(outputs[:, inputs.shape[1]:], skip_special_tokens=True)[0]
    
    return generated_text.strip()

iface = gr.Interface(
    fn=generate_text,
    inputs=[
        gr.Textbox(lines=3, label="Input Prompt"),
        gr.Textbox(lines=2, label="System Message", value="あなたは誠実で優秀な日本人のアシスタントです。"),
    ],
    outputs=gr.Textbox(label="Generated Text"),
    title="Llama-3.1-Swallow Text Generation",
    description="Enter a prompt and optional system message to generate text using the Llama-3.1-Swallow model. This model is optimized for Japanese language input and output.",
)

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