app.py

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

BASE_MODEL = "Qwen/Qwen2.5-3B-Instruct"
LORA_REPO = "Doanlol/qwen25-3b-van-lora"

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

base_model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto" if torch.cuda.is_available() else None,
    trust_remote_code=True,
)

model = PeftModel.from_pretrained(base_model, LORA_REPO)
model.eval()

SYSTEM_PROMPT = "Bạn là trợ lý viết văn tiếng Việt, lập luận rõ ràng, cảm xúc, đúng trọng tâm đề."

def generate_essay(prompt, max_new_tokens, temperature, top_p):
    messages = [
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": prompt},
    ]
    text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    inputs = tokenizer(text, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            top_p=top_p,
            do_sample=True,
            repetition_penalty=1.05,
            eos_token_id=tokenizer.eos_token_id,
        )

    decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
    answer = decoded[len(text):].strip() if decoded.startswith(text) else decoded
    return answer

demo = gr.Interface(
    fn=generate_essay,
    inputs=[
        gr.Textbox(lines=8, label="Nhập đề văn / yêu cầu"),
        gr.Slider(128, 1024, value=512, step=32, label="max_new_tokens"),
        gr.Slider(0.1, 1.5, value=0.7, step=0.1, label="temperature"),
        gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="top_p"),
    ],
    outputs=gr.Textbox(lines=16, label="Bài làm"),
    title="Qwen2.5-3B Văn AI (LoRA)",
    description="Sinh bài văn tiếng Việt từ model LoRA đã fine-tune.",
)

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