Skip to content
Robotics

Shallow-π0.5 L6

компактный ученик для SO-101

1 min read

Это 6-слойный студент π0.5, дистиллированный из 18-слойного учителя для роборуки SO-101. Задача — взять красный блок и положить его на цель с X. У студента 1.749 B параметров против 3.353 B у учителя; выбран чекпоинт checkpoints/6000.

Ключевые особенности

  • В VLM PaliGemma gemma_2b и action expert gemma_300m по 6 слоёв вместо 18; vision SigLIP So400m/14 — 27 слоёв, скопирован и дообучен.
  • Loss = task loss + velocity distillation; attention-distillation из статьи не используется.
  • Чекпоинты OpenPI (JAX/Orbax) на обновлениях 2000–10000; выбран 6000 по минимальному расхождению с учителем на шагах 0–9.
  • На H100 end-to-end P50 30.2 мс против 51.1 мс у учителя; на A100 — 47.2 против 86.1 мс.
  • Физических роллаутов пока нет: протокол 20 раскладок × 2 политики подготовлен, результатов нет.

Пример

huggingface-cli download jiajinda/so101-shallow-pi05-l6-red-block-to-target-v234 --revision <rev> --include "checkpoints/6000/*" "code/*" "examples/*" --local-dir so101-shallow
cd so101-shallow
python code/serve_student.py --checkpoint checkpoints/6000 --selftest_npz examples/selftest_obs.npz
ПодробнееУчитель π0.5 — 3.353 B, студент — 1.749 B; выбран checkpoints/6000.
Latency P50: H100 — 30.2 мс end-to-end, A100 — 47.2 мс; у учителя 51.1 и 86.1 мс.
Обучение: 10 000 шагов AdamW, 1× H100 80GB, пик 50.4 GB, ≈4.9 ч.
Нужен OpenPI commit 215abfb и папка code/: штатный конфиг π0.5 чекпоинты не загрузит.
Физических испытаний пока нет.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.7 GB · ready files4.6 GB
  • FP8 / INT8recommendedweights 1.9 GB2.6 GB
  • 4-bitweights 1.1 GB1.7 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 · BF16 — 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.

    # jiajinda/so101-shallow-pi05-l6-red-block-to-target-v234
    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 jiajinda/so101-shallow-pi05-l6-red-block-to-target-v234 --local-dir ~/models/so101-shallow-pi05-l6-red-block-to-target-v234
    # 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 so101-03-pi05-red-block-to-target-v234 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 1All on Hugging Face

About the model

jiajinda/so101-shallow-pi05-l6-red-block-to-target-v234
Author
jiajinda
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