Skip to content
Robotics

bento_ur7e_fineart_v2

FineART-VLA для UR7e: план на язы��е, действия flow-matching

2 min read

Это файнтюн политики FineART-VLA (метод от Scale AI и Hugging Face) для робота ur7e. Модель на 4.14B параметров предсказывает следующую подзадачу на языке и выполняет её flow-matching action expert, а knowledge insulation защищает предобученный VLM при дообучении.

Ключевые особенности

  • Двухуровневая схема: языковая голова планирует следующую подзадачу (System 2), action expert исполняет её (System 1).
  • Обучена под ur7e с камерами top и wrist: на входе состояние (14,) и три изображения 3×360×640, на выходе 7-мерное действие.
  • Датасет bento_ur7e_v5: 377 эпизодов, 767 361 кадр при 60 FPS; задачи — собрать ланчбокс с курицей и брокколи или с рыбой и сосисками.
  • 80 000 шагов обучения, batch size 16, AdamW, learning rate 5e-5, LeRobot 0.6.2.
  • Knowledge insulation защищает предобученный VLM от деградации во время файнтюна.

Пример

Фрагмент запуска политики на роботе; имена камер в --robot.cameras должны совпадать с top и wrist.

lerobot-rollout \
  --strategy.type=base \
  --robot.type=ur7e \
  --robot.port=<your_robot_port> \
  --policy.path=polarisai-robots/bento_ur7e_fineart_v2 \
  --task="Pack me a lunchbox with two fried chicken pieces and two brocolli." \
  --duration=60
ПодробнееБенчмарков нет: карточка не приводит результаты оценки.
Параметров — 4.14B; лицензия — gemma.
Дообучение: 80k шагов, batch 16, AdamW, lr 5e-5, seed 1000, LeRobot 0.6.2.
Ограничение: политика обучена под конкретного робота ur7e и задачи сборки ланчбокса; перенос на другие сценарии не описан.
Discuss Telegram 0 24

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.7 GB · ready files10 GB
  • FP8 / INT8recommendedweights 3.9 GB4.7 GB
  • 4-bitweights 2.2 GB2.9 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 2080 Ti

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.

    # polarisai-robots/bento_ur7e_fineart_v2
    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 polarisai-robots/bento_ur7e_fineart_v2 --local-dir ~/models/bento_ur7e_fineart_v2
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to bento_ur7e_fineart_v2
  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. paligemma-3b-pt-224family root
  2. pi05_base
  3. FineART-VLA-subtask-midtrained
  4. bento_ur7e_fineart_v2

New posts about versions of paligemma-3b-pt-224 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

polarisai-robots/bento_ur7e_fineart_v2
Author
polarisai-robots
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