Skip to content
Robotics

Cosmos3 Edge

SO-101 soarm_v4 — LoRA для захвата пробирки

1 min readupdated

Это LoRA-адаптер к nvidia/Cosmos3-Edge для робота SO-101. Он обучен на 100 демонстрациях (39 280 кадров при 30 FPS) брать пробирку и класть её в коробку. В репозитории лежат адаптер, обученные action-модули, статистика нормализации, процессоры и токенизатор; для загрузки нужна базовая модель.

Ключевые особенности

  • Политика LeRobot cosmos3_edge в режиме WAM с PEFT-aware загрузкой.
  • Управление по 6 суставам: shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper.
  • Две камеры: observation.images.front и observation.images.wrist; RGB 848×480, на входе 256×256 на вид.
  • Action chunk из 16 шагов на 30 Hz; углы суставов в градусах, gripper 0–100.
  • Чекпойнт шага 20 000; инференс по умолчанию в bfloat16 с 10 шагами денойзинга.

Пример

Задача для политики: Grab the test tube into the box. Подайте изображения с фронтальной и wrist-камеры, состояние в указанном порядке суставов и калибровку, соответствующую сбору данных. На выходе — 16 действий с частотой 30 Hz.

ПодробнееЧекпойнт: шаг 20 000. Исходное обучение записывало 30 шагов инференса, экспорт использует 10 без изменения весов.
Нужен доступ к nvidia/Cosmos3-Edge на его условиях; отдельно базовые веса не загружаются.
Систематических замеров успеха на реальном роботе нет; скорость зависит от железа.
Discuss Telegram 0 19

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 7.2 GB · ready files8.5 GB
  • FP8 / INT8recommendedweights 3.6 GB4.5 GB
  • 4-bitweights 2.1 GB2.8 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 · BF16 — from RTX 3080

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

This is a LoRA: counted together with its base model Cosmos3-Edge.

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.

    # qixunxiehuangbao/cosmos3_edge_soarm_v4_020000
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 nvidia/Cosmos3-Edge --local-dir ~/models/Cosmos3-Edge
    hf download qixunxiehuangbao/cosmos3_edge_soarm_v4_020000 --local-dir ~/models/cosmos3_edge_soarm_v4_020000
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Cosmos3-Edge
  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

Requires the base model

nvidia/Cosmos3-Edge

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of Cosmos3-Edge (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 8All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

qixunxiehuangbao/cosmos3_edge_soarm_v4_020000
Author
qixunxiehuangbao
Category
Robotics
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