Skip to content
Robotics

SmolVLA SO-101 M6

политика вставки штифта для SO-101

2 min readupdated

Это файнтюн SmolVLA на 0,45 млрд параметров от dhruv1672 на базе lerobot/smolvla_base для робота SO-101. Политика учится вставлять штифт 20 мм в глухое отверстие с радиальным зазором 0,5–2,5 мм в Isaac Sim 6.0.1 / Isaac Lab 3.0. В репозитории только файлы инференса, без состояния обучения.

Ключевые особенности

  • Двухэтапный файнтюн: M5 на 80 pan-shaped эпизодах (72 в обучении), затем M6 на 40 записях recovery-v2 — 80 чистых сегментов и 17 152 кадра.
  • Обучены action expert, проекции и все 12 блоков SigLIP; языковая модель заморожена.
  • Управление 30 Гц: 5 суставов руки в градусах + гриппер 0..100; камеры overhead и wrist 480x640.
  • Действия относительны состояния суставов на старте чанка; для замкнутого контура нужен серверный путь проекта.
  • Опциональные серверные фильтры blend, Kalman и cap без изменения весов.

Пример

from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy
policy = SmolVLAPolicy.from_pretrained(
    'dhruv1672/smolvla-so101-peg-M6-recovery-vision-trained',
    revision='M6-ckpt-020000',
)
ПодробнееЗакрытый контур, 80 эпизодов: 23/80 без фильтра, 29/80 с blend, 30/80 с Kalman, 24/80 с cap.
Обучение M6 заняло около 2 ч на RTX 5080 Laptop 16 ГБ при ~10 ГБ VRAM.
Это только симуляция без проверки на реальном роботе; успех сильно ниже скриптового эксперта (30/30 на 0,5–1,0 мм), цель 99/100 не достигнута.
Одна задача и один промпт; обобщение на язык и цвет не обучалось.
Discuss Telegram 0 12

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 1.3 GB · ready files1.9 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.

    # dhruv1672/smolvla-so101-peg-M6-recovery-vision-trained
    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 dhruv1672/smolvla-so101-peg-M6-recovery-vision-trained --local-dir ~/models/smolvla-so101-peg-M6-recovery-vision-trained
    # 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. smolvla-so101-peg-M6-recovery-vision-trained

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

dhruv1672/smolvla-so101-peg-M6-recovery-vision-trained
Author
dhruv1672
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

120

Ask AI about this model

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

Comments