Skip to content
Robotics

Beverage Multi-Task DiT

чекпоинт для перемешивания с исправленной нормализацией

1 min read

Это inference-экспорт политики multi_task_dit из LeRobot 0.6.1 от 2chan: около 0,28 млрд параметров, обучение 10 000 шагов только на задаче Stir the mixture in the cup with the muddler. Архитектура поддерживает несколько задач, но данный чекпоинт обучен лишь для перемешивания; это не тренировочный чекпоинт.

Ключевые особенности

  • LeRobot 0.6.1 и multi_task_dit; state и action — 12 измерений: левая рука, затем правая.
  • Исправлена нормализация левого gripper (state index 5): диапазон расширен до 100, исходное значение по-прежнему маппится в -1; веса и другие размерности не менялись.
  • Вход: три RGB-камеры left_wrist, right_wrist, top по 3x480x640; история — 2 кадра.
  • Выход: абсолютные целевые позиции суставов, action horizon 32, очередь — 24 шага.
  • Суставы задаются в градусах, gripper — в процентной шкале; при деплое нужна та же калибровка.
  • Нужны CLIP-энкодеры и токенизатор openai/clip-vit-base-patch16.

Пример

from lerobot.policies.multi_task_dit.modeling_multi_task_dit import MultiTaskDiTPolicy
from lerobot.policies.factory import make_pre_post_processors

repo_id = "2chan/beverage-multi-task-dit-20261105-10k"
policy = MultiTaskDiTPolicy.from_pretrained(repo_id).eval()
preprocessor, postprocessor = make_pre_post_processors(
    policy.config, pretrained_path=repo_id
)
policy.reset()
ПодробнееОффлайн-тест: возмущение gripper 0,001 меняло одну команду руки примерно на 179,77°; после ремонта — менее 0,00001°.
Нужен LeRobot 0.6.1 и CLIP-энкодеры/токенизатор openai/clip-vit-base-patch16.
Аппаратное исполнение не проверялось; ошибки трекинга остаются; это inference-экспорт, не resumable training checkpoint.
Discuss Telegram 0 17

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 1.0 GB · ready files1.6 GB
  • BF16weights 0.5 GB1.1 GB
  • FP8 / INT8recommendedweights 0.3 GB0.8 GB
  • 4-bitweights 0.1 GB0.7 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.

    # 2chan/beverage-multi-task-dit-20261105-10k
    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 2chan/beverage-multi-task-dit-20261105-10k --local-dir ~/models/beverage-multi-task-dit-20261105-10k
    # 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 beverage-multi-task-dit-20261105-10k (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

2chan/beverage-multi-task-dit-20261105-10k
Author
2chan
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

170

Ask AI about this model

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

Comments

Robotics

gaze_act

gaze_act