mini-Ichigo-llama3.2-3B-s-instruct

Model Details

We have developed and released the family Ichigo-llama3s. This family is natively understanding audio and text input.

We expand the Semantic tokens experiment with WhisperVQ as a tokenizer for audio files from homebrewltd/mini-Ichigo-llama3.2-3B-s-base with nearly 1B tokens from Instruction Speech WhisperVQ v3 dataset.

Model developers Homebrew Research.

Input Text and sound.

Output Text.

Model Architecture Llama-3.

Language(s): English.

Intended Use

Intended Use Cases This family is primarily intended for research applications. This version aims to further improve the LLM on sound understanding capabilities.

Out-of-scope The use of llama3-s in any manner that violates applicable laws or regulations is strictly prohibited.

How to Get Started with the Model

Try this model using Google Colab Notebook.

First, we need to convert the audio file to sound tokens

device = "cuda" if torch.cuda.is_available() else "cpu"
if not os.path.exists("whisper-vq-stoks-medium-en+pl-fixed.model"):
    hf_hub_download(
        repo_id="jan-hq/WhisperVQ",
        filename="whisper-vq-stoks-medium-en+pl-fixed.model",
        local_dir=".",
    )
vq_model = RQBottleneckTransformer.load_model(
        "whisper-vq-stoks-medium-en+pl-fixed.model"
    ).to(device)
vq_model.ensure_whisper(device)
def audio_to_sound_tokens(audio_path, target_bandwidth=1.5, device=device):

    wav, sr = torchaudio.load(audio_path)
    if sr != 16000:
        wav = torchaudio.functional.resample(wav, sr, 16000)
    with torch.no_grad():
        codes = vq_model.encode_audio(wav.to(device))
        codes = codes[0].cpu().tolist()

    result = ''.join(f'<|sound_{num:04d}|>' for num in codes)
    return f'<|sound_start|>{result}<|sound_end|>'

Then, we can inference the model the same as any other LLM.

def setup_pipeline(model_path, use_4bit=False, use_8bit=False):
    tokenizer = AutoTokenizer.from_pretrained(model_path)

    model_kwargs = {"device_map": "auto"}

    if use_4bit:
        model_kwargs["quantization_config"] = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_compute_dtype=torch.bfloat16,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type="nf4",
        )
    elif use_8bit:
        model_kwargs["quantization_config"] = BitsAndBytesConfig(
            load_in_8bit=True,
            bnb_8bit_compute_dtype=torch.bfloat16,
            bnb_8bit_use_double_quant=True,
        )
    else:
        model_kwargs["torch_dtype"] = torch.bfloat16

    model = AutoModelForCausalLM.from_pretrained(model_path, **model_kwargs)

    return pipeline("text-generation", model=model, tokenizer=tokenizer)

def generate_text(pipe, messages, max_new_tokens=64, temperature=0.0, do_sample=False):
    generation_args = {
        "max_new_tokens": max_new_tokens,
        "return_full_text": False,
        "temperature": temperature,
        "do_sample": do_sample,
    }

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

# Usage
llm_path = "homebrewltd/llama3.1-s-instruct-v0.2"
pipe = setup_pipeline(llm_path, use_8bit=True)

Training process

Training Metrics Image: Below is a snapshot of the training loss curve visualized.

MMLU:

AudioBench Eval:

Hardware

GPU Configuration: Cluster of 10x NVIDIA A6000-48GB.

GPU Usage:

• Fine-tuning: 12 hours.

Training Arguments

We utilize torchtune library for the latest FSDP2 training code implementation.

Examples

1. Good example:

2. Misunderstanding example:

3. Off-tracked example:

Citation Information

BibTeX:

@article{Llama3-S: Sound Instruction Language Model 2024,
  title={Llama3-S},
  author={Homebrew Research},
  year=2024,
  month=August},
  url={https://huggingface.co/homebrewltd/llama3.1-s-2024-08-20}

Acknowledgement

• WhisperSpeech

• Meta-Llama-3.1-8B-Instruct