Skip to content
Robotics

GR00T N1.7 rotate_qrcode

VLA-политика для RoboTwin 2.0

2 min read

Файнтюн NVIDIA Isaac GR00T N1.7-3B на симуляторных данных RoboTwin 2.0 для задачи rotate_qrcode. Отличие от предыдущего v1-чекпойнта — только обучающие данные: к 50 чистым демонстрациям добавлена аугментация v9clut, сгенерированная через Cosmos3-Nano-Sim-Transfer. Архитектура та же: vision-language бэкбон Cosmos-Reason2-2B заморожен, обучается flow-matching DiT action head.

Ключевые особенности

  • Embodiment NEW_EMBODIMENT с robotwin_aloha_config.py: 6 суставов руки (relative) + абсолютный gripper на каждую руку.
  • Три камеры: cam_high, cam_left_wrist, cam_right_wrist; горизонт действий — 16.
  • 10k шагов обучения, глобальный батч 64, 2 GPU NVIDIA B300, чекпойнт каждые 2.5k.
  • Релизный checkpoint-10000 отобран на отдельном валидационном seed (Hard, 50 эпизодов).
  • Аугментация даёт кратный рост: Easy/Hard выросли с 49/13 у v1 до 56/36.

Пример

Оценка идёт через eval_policy.py из RoboTwin 2.0 с GR00T policy server: на каждый инференс выполняется 8 шагов действий, инструкции — unseen, seed 0, по 100 эпизодов на режим. Полный рецепт и лог экспериментов — в репозитории nota-github/xpu-gr00t-aug-recipe.

ПодробнееSuccess rate (100 эпизодов): Easy 56% / Hard 36% против 14% / 3% у baseline и 49% / 13% у v1.
McNemar на тех же 100 сценах: Hard +24, p<0.001.
Данные: RoboTwin 2.0 (MIT) и видео Cosmos3-Nano-Sim-Transfer (OpenMDW 1.1).
Лицензия — NVIDIA Open Model License (производная от GR00T N1.7).
Обучение потребовало 2 GPU NVIDIA B300; ограничение — только задача rotate_qrcode в RoboTwin 2.0.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 6.4 GB · ready files7.6 GB
  • FP8 / INT8recommendedweights 3.2 GB4.0 GB
  • 4-bitweights 1.8 GB2.5 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 5060 Ti 8GB

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.

    # wjwjwjwj/gr00t-n1.7-robotwin-rotate_qrcode-simcosmos-aug
    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 wjwjwjwj/gr00t-n1.7-robotwin-rotate_qrcode-simcosmos-aug --local-dir ~/models/gr00t-n1.7-robotwin-rotate_qrcode-simcosmos-aug
    # 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 GR00T-N1.7-3B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 3All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face

About the model

wjwjwjwj/gr00t-n1.7-robotwin-rotate_qrcode-simcosmos-aug
Author
wjwjwjwj
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments