Skip to content
Audio

mixit-audioset-fuss-sparsity8-ft2k

дообученный MixIT-разделитель на четыре источника

1 min read

Это четырёхвыходной разделитель MixIT, дообученный за 2 000 optimizer updates от teacher sammyliu/mixit-audioset-fuss-step970k. Сохраняет архитектуру real-mask TDCN++ на 9 223 176 параметров и добавляет L1/L2 source-sparsity penalty с весом 8. Обучение шло только на неразмеченных смесях AudioSet.

Ключевые особенности

  • Четыре анонимных неупорядоченных выхода; вход 16 kHz.
  • L1/L2 source-sparsity penalty с весом 8; hard MixIT с negative thresholded SNR и потолком 30 dB.
  • 32 блока TDCN++, 256 analysis filters, width 256/512, encoder window/hop 40/20.
  • Дообучение: 2 000 optimizer updates; выбор по validation, сохраняется финальный checkpoint.
  • FUSS test: 1S SI-SDR 12.127 против 10.466 у базы; multi-source SI-SDRi 11.016 против 10.856.

Пример

from mixit_hf import MixITModel
separator = MixITModel.from_pretrained(
    repo_id="sammyliu/mixit-audioset-fuss-sparsity8-ft2k",
    checkpoint_filename="model.pt",
    device="cuda",
)
result = separator.separate_file("mixture_16khz.wav")
separator.write_stems(result, "separated")
ПодробнееFUSS test: 1S SI-SDR 12.127, multi-source SI-SDRi 11.016, correct source-count fraction 0.380.
Held-out gain над базой +0.161 dB, 95% интервал [-0.061, +0.383] — превосходство по multi-source не установлено.
Инференс: Python 3.10+, CUDA или CPU, файлы длиннее 10 с требуют чанкинга; веса и артефакты — proprietary по лицензии Samuel Liu.
Обучение: два A100, global batch 256, BF16/TF32; 2 000 updates.
Discuss Telegram 0 31

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # sammyliu/mixit-audioset-fuss-sparsity8-ft2k
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download sammyliu/mixit-audioset-fuss-sparsity8-ft2k --local-dir ~/models/mixit-audioset-fuss-sparsity8-ft2k
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of mixit-audioset-fuss-sparsity8-ft2k (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

sammyliu/mixit-audioset-fuss-sparsity8-ft2k
Author
sammyliu
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

310

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments