app.py

# import gradio as gr
# from huggingface_hub import InferenceClient
# import os

# client = InferenceClient(
#     model="mistralai/Mistral-Small-24B-Instruct-2501",
#     token=os.getenv('HF_TOKEN')
# )

# def chat_fn(message, system_message, history_str, max_tokens, temperature, top_p):
#     # Convert history string (optional) to message list
#     messages = [{"role": "system", "content": system_message}]
    
#     if history_str:
#         # Format: user1||assistant1\nuser2||assistant2
#         for pair in history_str.split("\n"):
#             if "||" in pair:
#                 user_msg, assistant_msg = pair.split("||", 1)
#                 messages.append({"role": "user", "content": user_msg})
#                 messages.append({"role": "assistant", "content": assistant_msg})
    
#     messages.append({"role": "user", "content": message})

#     # Get response from HF
#     response = ""
#     for chunk in client.chat_completion(
#         messages=messages,
#         stream=True,
#         max_tokens=max_tokens,
#         temperature=temperature,
#         top_p=top_p,
#     ):
#         response += chunk.choices[0].delta.content or ""
    
#     return response

# demo = gr.Interface(
#     fn=chat_fn,
#     inputs=[
#         gr.Textbox(lines=2, label="User Message"),
#         gr.Textbox(value="You are a friendly Chatbot.", label="System Prompt"),
#         gr.Textbox(lines=4, placeholder="user||bot\nuser2||bot2", label="Conversation History (optional)"),
#         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max Tokens"),
#         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
#         gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
#     ],
#     outputs="text",
#     allow_flagging="never",
#     title="LLM Budaya",
#     description="Chatbot menggunakan model HuggingFace Zephyr-7B"
# )

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

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

# Load model & tokenizer
model_id = "mistralai/Mistral-Small-24B-Instruct-2501"
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Load model di CPU
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float32,
    device_map={"": "cpu"}
)

# Inference function
def chat_fn(message, system_prompt, max_tokens, temperature, top_p):
    prompt = f"<s>[INST] {system_prompt.strip()}\n{message.strip()} [/INST]"

    inputs = tokenizer(prompt, return_tensors="pt").to("cpu")

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

    decoded = tokenizer.decode(output[0], skip_special_tokens=True)
    return decoded.split("[/INST]")[-1].strip()

# Gradio UI
demo = gr.Interface(
    fn=chat_fn,
    inputs=[
        gr.Textbox(lines=2, label="User Message"),
        gr.Textbox(value="You are a helpful and concise assistant.", label="System Prompt"),
        gr.Slider(minimum=1, maximum=1024, value=256, step=1, label="Max Tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
    ],
    outputs="text",
    title="Mistral-Small-24B CPU Chat",
    description="Chatbot menggunakan model Mistral-Small-24B-Instruct-2501 dijalankan lokal via CPU. Ini akan berjalan lambat.",
    flagging_mode="never",
)

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