Skip to content
Robotics

Jocasta-diff-VLA

адаптация Dream-VLA для манипуляций Panda

2 min read

Jocasta-diff-VLA — inference-чекпоинт от RobotLearningLab на базе Dream-VLA-7B для восьми семейств симулированных задач манипуляции Panda в robosuite. Модель использует диффузионный языковой бэкбон и непрерывную Flow Matching action head от авторов Dream-VLX. Одна совместно обученная политика набрала 332/512 (64.8%) успехов на фиксированной development-выборке.

Ключевые особенности

  • Один policy для 8 семейств задач: Lift, PickPlace, Stack, Door, NutAssembly, ToolHang, Push и Drawer.
  • В репозитории полный inference-чекпоинт: LoRA, обученная action head, проекторы шумных действий и проприоцепции, нормализация датасета, processor и конфиг обучения.
  • На входе две RGB-камеры 224×224, 8D проприоцепция и текстовая инструкция; на выходе — chunk из 8 действий по 7 компонентов.
  • Инференс: 4 шага Flow Matching, BF16/SDPA, одна видимая CUDA GPU; в симуляции CUDA-инференс и EGL-рендеринг лучше держать на разных физических GPU.
  • База зафиксирована на Dream-org/Dream-VLA-7B@0cd0e25c1771c5b468c74585b5cf35c4a4aa315c; исходные веса нужны отдельно, а AutoModel.from_pretrained не восстановит обученную политику.

Пример

from load_policy import load_policy, predict_action
policy, processor, cfg = load_policy('RobotLearningLab/Jocasta-diff-VLA')
# primary, wrist: uint8 RGB (224,224,3); proprio: float32 (8,)
actions = predict_action(policy, processor, cfg, primary, wrist, proprio,
                         'pick up the red cube', seed=84907)
# actions.shape == (8, 7)
ПодробнееРезультат: 332/512 (64.8%) на фиксированной dev-выборке; на PickPlace — 40/64 против 0/64 у исходной Dream-VLA.
Слабое место — ToolHang: 9/64 (14.1%); другие роботы, реальное железо и reserved test split не проверялись.
Обучение: 8,140 успешных траекторий, 4,841,083 шага, 100,000 обновлений, LoRA rank 32.
Референсный инференс: Python 3.10.18, PyTorch 2.5.1+cu124, Transformers 4.51.3, BF16/SDPA на H200; одна CUDA GPU.
Это development-результаты на задачах robosuite/Panda, не официальные LIBERO scores; для остальных семи задач baseline не заявлен.
Discuss Telegram 11

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.

    # RobotLearningLab/Jocasta-diff-VLA
    # ===== 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 RobotLearningLab/Jocasta-diff-VLA --local-dir ~/models/Jocasta-diff-VLA
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Jocasta-diff-VLA
  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 Jocasta-diff-VLA (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

RobotLearningLab/Jocasta-diff-VLA
Author
RobotLearningLab
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

011

Ask AI about this model

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

Comments