th3james/assembly-decoder

1import sys2from threading import Thread3 4import gradio as gr5import spaces6from transformers import (7    AutoModelForCausalLM,8    AutoTokenizer,9    TextIteratorStreamer,10    BitsAndBytesConfig,11)12 13 14import torch15 16 17MODEL = "microsoft/Phi-3.5-mini-instruct"18 19if torch.cuda.is_available():20    device = "cuda"21elif sys.platform == "darwin" and torch.backends.mps.is_available():22    device = "mps"23else:24    device = "cpu"25 26 27# TODO understand this28quantization_config = BitsAndBytesConfig(29    load_in_4bit=True,30    bnb_4bit_compute_dtype=torch.bfloat16,31    bnb_4bit_use_double_quant=True,32    bnb_4bit_quant_type="nf4",33)34 35tokenizer = AutoTokenizer.from_pretrained(MODEL)36model = AutoModelForCausalLM.from_pretrained(37    MODEL,38    torch_dtype=torch.bfloat16,39    device_map="auto",40    quantization_config=quantization_config,41)42 43 44@spaces.GPU()45def stream_chat(46    message: str,47    history: list,48    system_prompt: str,49    temperature: float = 0.8,50    max_new_tokens: int = 1024,51    top_p: float = 1.0,52    top_k: int = 20,53    penalty: float = 1.2,54):55    print(f"message: {message}")56    print(f"history: {history}")57 58    conversation = [{"role": "system", "content": system_prompt}]59    for prompt, answer in history:60        conversation.extend(61            [62                {"role": "user", "content": prompt},63                {"role": "assistant", "content": answer},64            ]65        )66 67    conversation.append({"role": "user", "content": message})68 69    input_ids = tokenizer.apply_chat_template(70        conversation, add_generation_prompt=True, return_tensors="pt"71    ).to(model.device)72 73    streamer = TextIteratorStreamer(74        tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True75    )76 77    generate_kwargs = dict(78        input_ids=input_ids,79        max_new_tokens=max_new_tokens,80        do_sample=False if temperature == 0 else True,81        top_p=top_p,82        top_k=top_k,83        temperature=temperature,84        eos_token_id=[128001, 128008, 128009],85        streamer=streamer,86    )87 88    with torch.no_grad():89        thread = Thread(target=model.generate, kwargs=generate_kwargs)90        thread.start()91 92    buffer = ""93    for new_text in streamer:94        buffer += new_text95        yield buffer96 97 98"""99For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface100"""101demo = gr.ChatInterface(102    stream_chat,103    additional_inputs=[104        gr.Textbox(value="You are an ARM Assembly language decoder. You receive a line of Arm assembly and respond with a description of what the instruction does.", label="System message"),105        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),106        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),107        gr.Slider(108            minimum=0.1,109            maximum=1.0,110            value=0.95,111            step=0.05,112            label="Top-p (nucleus sampling)",113        ),114    ],115)116 117 118if __name__ == "__main__":119    demo.launch()120