Skip to content
Robotics

AlphaBot2 OVMM SFT3000

замороженный SFT-бейзлайн для OVMM

2 min read

Это публичная копия policy-чекпоинта manip_recovery_opt_v1/ckpt_step0003000/policy от Neptune-T на базе alphabot2/aibot2_grab_bag_combined_over_under_pi05. Модель на 4.14 млрд параметров, библиотека LeRobot, для управления роботом-манипулятором по состоянию, текстовой задаче и шести RGB-камерам. Это inference/fine-tuning экспорт, без нового обучения и конвертации весов.

Ключевые особенности

  • На входе — state16, текст задачи и шесть упорядоченных RGB-ключей: camera_left, camera_right, camera_top, camera_top_h5, camera_top_h10, camera_top_h15.
  • Нужен runtime LeRobot pi0.5 из проекта; версии зафиксированы в release/runtime.json, код serving лежит в release/code/scripts/alphabot2_pi05/.
  • В корне — полные FP32-веса policy, конфиг, pre/postprocessors, тензоры нормализации и tokenizer. Это экспорт для инференса и файнтюна, а не точный optimizer/RNG resume-чекпоинт.
  • В training/ — исходная конфигурация, summary и лог; в release/ — код контроллера, версии пакетов и свидетельства оценки. Абсолютные пути в метаданных нужно перелокализовать, но корневые веса самодостаточны.
  • Выход — 50-шаговый 18-мерный action chunk. История — реальные предыдущие кадры головы, а не повтор текущих.

Пример

from huggingface_hub import snapshot_download
model = snapshot_download('Neptune-T/alphabot2-ovmm-sft3000', local_dir='ovmm-model')
data = snapshot_download('Neptune-T/alphabot2-ovmm-sft', repo_type='dataset', local_dir='ovmm-data')
ПодробнееОбучение: продолжение 3000 шагов, global batch 12, старт из 250-шагового NAV-correction пилота; 80% phase-balanced original replay / 20% qualified NAV corrections.
Закрытый контур: TRAIN12 full success 4/12, development heldout 0/3, fixed development6 1/6 — только assisted simulator results с privileged-goal Planner и observable_placement_exit_v2, FP32, batch1.
Ограничения: не pure end-to-end navigation и не real-robot qualification; simulator assets/runtime не включены; исходные лицензии upstream сохраняются.
Discuss Telegram 0 30

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 15 GB · ready files17 GB
  • BF16weights 7.7 GB9.0 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 · 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.

    # Neptune-T/alphabot2-ovmm-sft3000
    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 Neptune-T/alphabot2-ovmm-sft3000 --local-dir ~/models/alphabot2-ovmm-sft3000
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to alphabot2-ovmm-sft3000
  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 aibot2_grab_bag_combined_over_under_pi05 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 1All on Hugging Face

About the model

Neptune-T/alphabot2-ovmm-sft3000
Author
Neptune-T
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

300

Ask AI about this model

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

Comments

Robotics

FlowPilot

FlowPilot