Skip to content
Robotics

Pi0.5 Task820

vision-only SFT для реального манипулятора

2 min read

Набор из трёх EMA-чекпоинтов (10k/20k/30k шагов) от Demomasterlqx — SFT-политика Pi0.5 для робототехнической задачи task820 firm-current. Двухветвевая LoRA и остальные обучаемые параметры уже влиты в полную базу Pi0.5, поэтому каждый step_* можно загружать отдельно без исходной модели и адаптера.

Ключевые особенности

  • Три независимых чекпоинта: step_10000, step_20000, step_30000; EMA 0.99, BF16, веса слиты с базой.
  • На входе только RGB основной и wrist-камеры плюс язык; нет state-токенов, силы, тактильных изображений и marker motion.
  • Архитектура Pi0.5: action horizon=50, внутренняя размерность действия 32, эффективный выход 13-мерный — 7 действий + 6 сил.
  • Обучение на xiangxin0923/realworld_task820_firm_current: 90 эпизодов, 13 446 кадров, 5 fps; промпты без наречий.
  • У силовых каналов loss-вес 0 — прямой супервизии нет; поле state нужно для нормализации, но не как вход модели.

Пример

from pathlib import Path
from huggingface_hub import snapshot_download
step = 30000
root = Path(snapshot_download(
    "Demomasterlqx/VTLA-RL-sft-lora-task820-firm-current-vision-nostate-force0",
    allow_patterns=[f"step_{step}/*"],
))
checkpoint = root / f"step_{step}"
ПодробнееБенчмарков и оценки успешности нет: проверка касается экспорта и структуры, не качества политики.
Запуск — в OpenPI PyTorch / T2-VLA, не через Transformers AutoModel; проверено в Docker 0.9.7.
Обучение: 2 GPU FSDP, micro batch 16, global batch 32, AdamW lr 2.5e-5, 30k шагов за 63 ч 18 мин.
Ограничения: только vision+language, без force/tactile/state-условия; силовые каналы без прямого обучения.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 6.7 GB · ready files7.9 GB
  • FP8 / INT8recommendedweights 3.4 GB4.2 GB
  • 4-bitweights 1.9 GB2.6 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 2060 · BF16 — from RTX 5060 Ti 8GB

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.

    # Demomasterlqx/VTLA-RL-sft-lora-task820-firm-current-vision-nostate-force0
    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 Demomasterlqx/VTLA-RL-sft-lora-task820-firm-current-vision-nostate-force0 --local-dir ~/models/VTLA-RL-sft-lora-task820-firm-current-vision-nostate-force0
    # 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 VTLA-RL-sft-lora-task820-firm-current-vision-nostate-force0 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Demomasterlqx/VTLA-RL-sft-lora-task820-firm-current-vision-nostate-force0
Author
Demomasterlqx
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