app.py

import os
import time
import spaces
import torch
import gradio as gr
from threading import Thread

HF_TOKEN = os.environ.get("HF_TOKEN", None)

TITLE = "<h1><center>Mistral-lab</center></h1>"

PLACEHOLDER = """
<center>
<p>Chat with Mistral AI LLM.</p>
</center>
"""

from huggingface_hub import snapshot_download
from pathlib import Path

mistral_models_path = Path.home().joinpath('mistral_models', '8B-Instruct')
mistral_models_path.mkdir(parents=True, exist_ok=True)

snapshot_download(repo_id="mistralai/Ministral-8B-Instruct-2410", allow_patterns=["params.json", "consolidated.safetensors", "tekken.json"], local_dir=mistral_models_path)

from mistral_inference.transformer import Transformer
from mistral_inference.generate import generate

from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from mistral_common.protocol.instruct.messages import AssistantMessage, UserMessage
from mistral_common.protocol.instruct.request import ChatCompletionRequest

device = "cuda" # for GPU usage or "cpu" for CPU usage

tokenizer = MistralTokenizer.from_file(f"{mistral_models_path}/tekken.json")
model = Transformer.from_folder(mistral_models_path)


@spaces.GPU()
def stream_chat(
    message: str, 
    history: list, 
    temperature: float = 0.3, 
    max_new_tokens: int = 1024, 
    top_p: float = 1.0, 
    top_k: int = 20, 
    penalty: float = 1.2,
):
    print(f'message: {message}')
    print(f'history: {history}')

    conversation = []
    for prompt, answer in history:
        conversation.append(UserMessage(content=prompt))
        conversation.append(AssistantMessage(content=answer))
    conversation.append(UserMessage(content=message))

    completion_request = ChatCompletionRequest(messages=conversation)
    
    tokens = tokenizer.encode_chat_completion(completion_request).tokens
    
    out_tokens, _ = generate(
        [tokens], 
        model, 
        max_tokens=max_new_tokens, 
        temperature=temperature,
        top_p = top_p,
        top_k = top_k,
        repetition_penalty=penalty,
        eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
    
    result = tokenizer.instruct_tokenizer.tokenizer.decode(out_tokens[0])
    
    return result
            
chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)

with gr.Blocks(theme="citrus") as demo:
    gr.HTML(TITLE)
    gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
    gr.ChatInterface(
        fn=stream_chat,
        chatbot=chatbot,
        fill_height=True,
        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
        additional_inputs=[
            gr.Slider(
                minimum=0,
                maximum=1,
                step=0.1,
                value=0.3,
                label="Temperature",
                render=False,
            ),
            gr.Slider(
                minimum=128,
                maximum=8192,
                step=1,
                value=1024,
                label="Max new tokens",
                render=False,
            ),
            gr.Slider(
                minimum=0.0,
                maximum=1.0,
                step=0.1,
                value=1.0,
                label="top_p",
                render=False,
            ),
            gr.Slider(
                minimum=1,
                maximum=20,
                step=1,
                value=20,
                label="top_k",
                render=False,
            ),
            gr.Slider(
                minimum=0.0,
                maximum=2.0,
                step=0.1,
                value=1.2,
                label="Repetition penalty",
                render=False,
            ),
        ],
        examples=[
            ["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
            ["What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter."],
            ["Tell me a random fun fact about the Roman Empire."],
            ["Show me a code snippet of a website's sticky header in CSS and JavaScript."],
        ],
        cache_examples=False,
    )


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