Skip to content
Robotics

MolmoAct2-YAM-StationeryBench-Sim

файнтюн для двуручных задач

1 min readupdated

Пилотный релиз xuxie1031: файнтюн allenai/MolmoAct2-BimanualYAM на две сложные двуручные задачи StationeryBench. Модель на 5,44 млрд параметров поднимает успех в физической симуляции с 0% до 36% и 76%.

Ключевые особенности

  • В симуляции: eraser_from_box — 36% (18/50), ruler_handover — 76% (38/50); базовая модель — 0/50 в обеих.
  • База — allenai/MolmoAct2-BimanualYAM; файлы, архитектура, processor и norm_stats.json те же.
  • 13 раундов: сначала только action expert, затем vision LoRA + LLM LoRA (r=64) + action expert.
  • Синтетические демонстрации в симуляции + публичные реальные данные YAM; датасеты двух задач выпущены.
  • Важно: действия нужно денормализовать без клиппинга в [-1, 1]; стандартный сервер клиппит и ухудшает модель.

Пример

git clone https://github.com/xuxie1031-droid/stationerybench-sim && cd stationerybench-sim
source scripts/env.sh && scripts/download.sh finetuned
scripts/repro_check.sh $GYRE_WORK/models/MolmoAct2-YAM-StationeryBench-Sim
ПодробнееБенчмарк: 18/50 (36%, 95% Wilson 24–50%) и 38/50 (76%, 63–86%) на sealed seeds; база — 0/50.
Обучение: ~263 L40S GPU-hours, 13 раундов; Ai2 сообщает ~2300 H100-hours для адаптации MolmoAct2 к YAM.
Ограничения: только симуляция, не тестировалось на реальном роботе; две задачи, по одной фиксированной сцене.
Лицензия Apache-2.0; результаты — на неофициальной реплике, не на роботе Robocurve.
Discuss Telegram 0 14

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 20 GB · ready files23 GB
  • BF16weights 10 GB12 GB
  • FP8 / INT8recommendedweights 5.1 GB6.1 GB
  • 4-bitweights 2.9 GB3.6 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 5060 Ti 8GB · F32 — from RTX 4090

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.

    # xuxie1031/MolmoAct2-YAM-StationeryBench-Sim
    # ===== 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 xuxie1031/MolmoAct2-YAM-StationeryBench-Sim --local-dir ~/models/MolmoAct2-YAM-StationeryBench-Sim
    # 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 MolmoAct2-BimanualYAM (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

xuxie1031/MolmoAct2-YAM-StationeryBench-Sim
Author
xuxie1031
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

140

Ask AI about this model

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

Comments