Skip to content
Other

LFPformer Large

эмбеддинги LFP для нейронауки

2 min readupdated

LFPformer Large — MaskPred — 35-тысячный Large-checkpoint от Shubhronyu из сравнения с matched decoder capacity (NeurIPS 2026). Модель инициализирована от facebook/wav2vec2-large и извлекает представления локальных потенциалов поля (LFP) из окон IBL. Это research-checkpoint с фиксированным бюджетом, а не финальная точка обучения.

Ключевые особенности

  • MaskPred — masked latent prediction без spatial auxiliary objective; база facebook/wav2vec2-large, 24 слоя, hidden 1024, 16 голов, ~317,5 млн параметров.
  • 35 000 шагов, effective global batch 512; обучен на IBL full-recording LFP windows.
  • Извлекает 1024-мерные эмбеддинги: helper ресемплит окна до 16 кГц без per-window нормализации и усредняет последние hidden states.
  • Вход — уже предобработанные одноканальные LFP-окна 3 с при 1250 Гц, массив (N, 3750).
  • Загружается через стандартный Wav2Vec2ForPreTraining; custom remote code не нужен.
  • Не ASR: не применяйте к LFP стандартную waveform-нормализацию audio processor.

Пример

pip install -r requirements.txt
python extract_embeddings.py --model . --input windows.npz --output embeddings.npz --device cpu
Подробнее35 000 шагов при effective global batch 512 — 17 920 000 предъявлений окон, но не столько уникальных записей.
Вход: окна 3 с при 1250 Гц, shape (N, 3750); helper ресемплит до 16 кГц без нормализации.
auxiliary XYZ head имел xyz_lambda=0 и не является обученным localization decoder; downstream-декодера в пакете нет.
Проверены строгая загрузка и smoke-check (1, 1024), но это не научная метрика. Лицензия весов не указана.
Discuss Telegram 0 19

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 1.2 GB · ready files1.8 GB
  • BF16weights 0.6 GB1.2 GB
  • FP8 / INT8recommendedweights 0.3 GB0.8 GB
  • 4-bitweights 0.2 GB0.7 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # Shubhronyu/LFPformer-large-MaskPred-35k
    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 Shubhronyu/LFPformer-large-MaskPred-35k --local-dir ~/models/LFPformer-large-MaskPred-35k
    # 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 wav2vec2-large (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face
Fine-tunes 12All on Hugging Face

About the model

Shubhronyu/LFPformer-large-MaskPred-35k
Author
Shubhronyu
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

190

Ask AI about this model

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

Comments