import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel, PeftConfig
import gradio as gr

model_repo = "nambn0321/LLM_model"

# Load LoRA adapter config 
peft_config = PeftConfig.from_pretrained(model_repo)

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=torch.float32
)

base_model = AutoModelForCausalLM.from_pretrained(
    peft_config.base_model_name_or_path,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
    offload_folder="./offload"
)

# Load adapter weights from your fine-tuned repo
model = PeftModel.from_pretrained(base_model, model_repo)

# Load tokenizer from the Hub repo
tokenizer = AutoTokenizer.from_pretrained(model_repo, use_fast=False)

def generate_response(prompt, max_tokens=128, temperature=0.7, top_p=0.9):
    try:
        chat = [{"role": "user", "content": prompt}]
        formatted_prompt = tokenizer.apply_chat_template(
            chat, tokenize=False, add_generation_prompt=True
        )

        inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)

        with torch.no_grad():
            outputs = model.generate(
                **inputs,
                max_new_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                do_sample=True,
                eos_token_id=tokenizer.eos_token_id,
                use_cache=False
            )

        decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)

        # Clean up output (optional)
        if "<|assistant|>" in decoded:
            decoded = decoded.split("<|assistant|>")[-1].strip()

        return decoded
    except Exception as e:
        return f"Error: {str(e)}"

iface = gr.Interface(
    fn=generate_response,
    inputs=[
        gr.Textbox(lines=4, label="Prompt"),
        gr.Slider(16, 512, value=128, step=16, label="Max Tokens"),
        gr.Slider(0.1, 1.5, value=0.7, label="Temperature"),
        gr.Slider(0.1, 1.0, value=0.9, label="Top-p")
    ],
    outputs="text",
    title="Fine-Tuned LLM",
    description="Interact with my fine-tuned LLM."
)

iface.launch()