Skip to content
Robotics

wam-openarm-vwam-pfaction-humvid-agv-50k-iter50000

OpenArm-политика, где action-поток слушает только pointflow

2 min read

Это экспериментальный чекпоинт VWAM для робота OpenArm (EXP-0019), вариант motion-pfaction-humvid. В основе — backbone Cosmos3-Nano MoT и ActionDiT с нуля; action-поток уменьшен до 4 блоков и читает K/V только pointflow-стадии, без K/V видео/VLM.

Ключевые особенности

  • ActionDiT уменьшен до 4 блоков (hidden 1024); блок k смотрит на свои action-строки и K/V pointflow-стадии k (per-block LayerNorm + проекция K/V 256→1024).
  • Pointflow: 12 точек кисти × 24 будущих смещения, 4 стадии, ширина 256; читает слои backbone 3/5/7/9.
  • Pointflow зашумляется под sigma видео/действия каждого сэмпла (общее расписание) и обучается flow-matching loss 0.2; action-loss доходит до backbone только через pointflow.
  • Данные motion-humvid: 6 робот-подмножеств + 4 человеческих подмножества как video-only строки (box source исключён), ровно 8 robot + 8 human на GPU, b64, 50k шагов, seed 42, LR 1e-4, warmup 2500, cosine до 10%.
  • Robot action 28-D: руки относительные, шея/кисти абсолютные; ActionDiT обучался с нуля.
  • Для инференса нужен pointflow-поток: текущие 12 точек кисти и совместный денойзинг flow и action; это не путь сервинга junhyeong.

Пример

Сценарий из карточки: на вход подаются текущие 12 точек кисти, flow и action денойзятся совместно. Штатный serving-путь junhyeong для этого чекпоинта не подходит.

ПодробнееЧекпоинт iter_000050000 — torch DCP с EMA, без оптимизатора; рядом scheduler/, trainer/, config.yaml/config.pkl и normalization.json с q01/q99.
Бенчмарков и требований к VRAM в карточке нет.
Инференс требует pointflow-поток и совместного денойзинга flow/action; это не штатный путь сервинга junhyeong.
Код: runtime junhyeong vwam-openarm-resume-20261103-v1 + pflow_action_l4.patch.
Discuss Telegram 0 10

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # jungwook2358/wam-openarm-vwam-pfaction-humvid-agv-50k-iter50000
    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 jungwook2358/wam-openarm-vwam-pfaction-humvid-agv-50k-iter50000 --local-dir ~/models/wam-openarm-vwam-pfaction-humvid-agv-50k-iter50000
    # 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 wam-openarm-vwam-pfaction-humvid-agv-50k-iter50000 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

jungwook2358/wam-openarm-vwam-pfaction-humvid-agv-50k-iter50000
Author
jungwook2358
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

100

Ask AI about this model

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

Comments