app.py

import gradio as gr
from huggingface_hub import login
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, TextIteratorStreamer
from threading import Thread
import torch

MODEL = "m-a-p/OpenCodeInterpreter-DS-33B"

system_message = "You are a computer programmer that can translate python code to C++ in order to improve performance"

def user_prompt_for(python):
    return f"Rewrite this python code to C++. You must search for the maximum performance. \
    Format your response in Markdown. This is the Code: \
    \n\n\
    {python}"

def messages_for(python):
    return [
        {"role": "system", "content": system_message},
        {"role": "user", "content": user_prompt_for(python)}
    ]

quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_quant_type="nf4"
)

tokenizer = AutoTokenizer.from_pretrained(MODEL)
model = AutoModelForCausalLM.from_pretrained(MODEL, device_map="auto", quantization_config=quant_config)
streamer = TextIteratorStreamer(tokenizer)

cplusplus = None
def translate(python):
    inputs = tokenizer.apply_chat_template(messages_for(python), return_tensors="pt").to(model.device)
    generation_kwargs = dict(
                            inputs,
                            streamer=streamer,
                            max_new_tokens=1024,
                            do_sample=False,
                            pad_token_id=tokenizer.eos_token_id,
                            eos_token_id=tokenizer.eos_token_id
                            )
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()
    cplusplus = ""
    for chunk in streamer:
        cplusplus += chunk
        yield cplusplus

demo = gr.Interface(fn=translate, inputs="code", outputs="markdown")
demo.launch()