Skip to content
Robotics

SO101 red/green SmolVLA v2

финальный чекпоинт для сортировки кубиков

1 min read

Это финальный inference-чекпоинт SmolVLA для робота SO101, обученный двум командам: положить зелёный или красный кубик в бумажную миску. Модель содержит 450 046 176 параметров, сохранена в FP32, базируется на lerobot/smolvla_base и использует библиотеку lerobot. Это ограниченный tabletop-эксперимент, а не универсальная робомодель.

Ключевые особенности

  • Fine-tune SmolVLA на 20 000 шагов: VLM заморожен, обучались expert/state projection.
  • Две инструкции: Put the green cube into the paper bowl и Put the red cube into the paper bowl.
  • SO101 state/action размерности 6; wrist и side RGB — 640×480 при 20 fps; policy переименовывает wrist→camera1, side→camera2.
  • Валидация на 23 145 кадрах: финальный flow loss green 0.4627, red 0.4275; это не real-world success rate.
  • Пройдены CPU strict reload и train-fitted processor-stat identity.

Пример

python prepare_runtime.py /path/to/downloaded-model /path/to/runtime-model
CHECK_ONLY=1 ./rollout_smolvla.sh eval-v2-preflight 1 /path/to/runtime-model
ПодробнееБенчмарки: на исторических раскладках C2/C3 — 11/12 (green 5/6, red 6/6); на novel-position screen — 9/16 (green 2/8, red 7/8).
Оценка получена offline visual review записанных видео, только 2 повтора на novel-условие; заявлений о широкой генерализации нет.
Нужен проект SmolVLA LeRobot 0.6.1 и FP32; лицензия для fine-tuned weights и записей не указана, условия upstream сохраняются.
Discuss Telegram 0 24

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 1.7 GB · ready files2.3 GB
  • BF16weights 0.8 GB1.4 GB
  • FP8 / INT8recommendedweights 0.4 GB1.0 GB
  • 4-bitweights 0.2 GB0.8 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 · 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.

    # zjumty/real-smolvla-red-green-v2-local-20261103
    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 zjumty/real-smolvla-red-green-v2-local-20261103 --local-dir ~/models/real-smolvla-red-green-v2-local-20261103
    # 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

  1. SmolLM2-360Mfamily root
  2. SmolLM2-360M-Instruct
  3. SmolVLM-500M-Instruct
  4. SmolVLM2-500M-Video-Instruct
  5. smolvla_base
  6. real-smolvla-red-green-v2-local-20261103

New posts about versions of SmolLM2-360M (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 8All on Hugging Face
Merges 1All on Hugging Face

About the model

zjumty/real-smolvla-red-green-v2-local-20261103
Author
zjumty
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

240

Ask AI about this model

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

Comments