app.py

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

# Setup model dan tokenizer
torch.random.manual_seed(0)
model = AutoModelForCausalLM.from_pretrained(
    "microsoft/Phi-3-mini-128k-instruct",
    device_map="cpu",  # Gunakan 'cpu' jika tidak ada GPU
    torch_dtype="auto",
    trust_remote_code=True,
    attn_implementation="eager"  # Menggunakan eager untuk menghindari masalah flash-attention
)

tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-128k-instruct")

# Pipeline untuk text-generation
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
)

# Fungsi untuk menghasilkan respons
def generate_response(input_text):
    messages = [
        {"role": "system", "content": "You are a helpful AI assistant."},
        {"role": "user", "content": input_text}
    ]
    
    generation_args = {
        "max_new_tokens": 500,
        "return_full_text": False,
        "temperature": 0.7,  # Bisa disesuaikan untuk variasi output
        "do_sample": True,   # Mengaktifkan sampling untuk variasi output
    }

    output = pipe(messages, **generation_args)
    return output[0]['generated_text']

# Membuat antarmuka menggunakan Gradio Blocks
with gr.Blocks() as demo:
    gr.Markdown("# AI Chatbot Assistant\nTanya apapun, saya siap membantu!")
    
    # Pesan pemberitahuan untuk penggunaan CPU
    gr.Markdown(
        "### ⚠ Sorry for the inconvenience. The Space is currently running on the CPU, which might affect performance. We appreciate your understanding."
    )
    
    # Tata letak output di atas input
    with gr.Row():
        output_box = gr.Textbox(
            label="AI Response", 
            placeholder="Respons akan muncul di sini...", 
            lines=10, 
            interactive=False  # Tidak dapat diisi manual
        )
    with gr.Row():
        input_box = gr.Textbox(label="Ask me anything!", placeholder="Tanyakan sesuatu...")
    with gr.Row():
        submit_button = gr.Button("Submit")
    
    # Aksi untuk submit
    submit_button.click(generate_response, inputs=input_box, outputs=output_box)

# Menjalankan antarmuka
demo.launch()