Skip to content
Robotics

StreamingWAM-DiT4DiT-LIBERO

streaming-адаптация для LIBERO

2 min readupdated

StreamingWAM-DiT4DiT-LIBERO — набор из двух чекпоинтов для роботизированных задач LIBERO. Это дополнительное обучение на DiT4DiT с видеобэкбоном Cosmos-Predict2.5-2B: сравниваются обычный Joint и streaming-вариант с историей наблюдений и действий. Код и команды запуска — в репозитории.

Ключевые особенности

  • Два чекпоинта: joint и streaming; оба стартуют с одной официальной инициализации DiT4DiT-LIBERO.
  • Streaming использует пять реальных визуальных наблюдений и восемь выполненных действий, уточняет буфер на восемь действий и исполняет четыре действия на replan.
  • Joint сохраняет одно текущее изображение и четыре обновления действий на replan; оба варианта используют парные виды основной и wrist-камеры.
  • В комплекте полный inference state dict (~21,3 ГБ), конфиги, статистика нормализации и метаданные; optimizer states и логи не включены.
  • Это чекпоинты для LIBERO simulation, не real-robot или RoboTwin.

Пример

git clone https://github.com/renshaojie233/StreamingWAM.git
cd StreamingWAM/extensions/dit4dit
python scripts/download.py --method both --output checkpoints/dit4dit
CUDA_VISIBLE_DEVICES=0 python scripts/smoke_inference.py --method streaming --checkpoint-dir checkpoints/dit4dit/streaming --cosmos-path checkpoints/cosmos
ПодробнееLIBERO-Plus raw-prompt success: joint — 70,88%, streaming — 76,37% (10 030 эпизодов, seed 7).
Каждый чекпоинт — около 21,3 ГБ; нужен отдельный DiT4DiT-окружение и pinned Diffusers, это не generic AutoModel.
Задержка: 135,526 мс у обычного варианта и 153,685 мс у streaming; end-to-end ускорение не заявляют.
Только LIBERO simulation; streaming-точка выбиралась по 200-entry screen, поэтому полная оценка не полностью held out.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 20 GB · ready files22 GB
  • FP8 / INT8recommendedweights 9.9 GB11 GB
  • 4-bitweights 5.6 GB6.6 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 5070 · BF16 — 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.

    # rsj2003/StreamingWAM-DiT4DiT-LIBERO
    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 rsj2003/StreamingWAM-DiT4DiT-LIBERO --local-dir ~/models/StreamingWAM-DiT4DiT-LIBERO
    # 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 dit4dit-model (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 1All on Hugging Face

About the model

rsj2003/StreamingWAM-DiT4DiT-LIBERO
Author
rsj2003
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