Skip to content
Robotics

ReTime / OPTR

чекпоинты Florence–Bridge и Qwen–Bridge для воспроизводимых seed-экспериментов

2 min readupdated

Релиз mmmmmcyaaa для robotics: server-authoritative код Florence–Bridge OPTR, полное 20K common-start состояние, снимки Uniform/Residual-OPTR на 20.5K, а также Qwen3-VL-4B–Bridge 29K common start и контракт окружения/seed для парных multi-seed экспериментов. Пакет на PyTorch включает конфиги, скрипты, манифесты и переносимый код обучения и оценки.

Ключевые особенности

  • Полный resumable state Florence 20K: model, optimizer, scheduler metadata и per-rank RNG — с него стартуют новые continuation-ветки.
  • Model-only снимки Uniform и Residual-OPTR на 20.5K — подходят для оценки, но не для точного продолжения optimizer/RNG.
  • Замороженный Residual proposal K=8, alpha=0.2 и манифесты provenance/фингерпринтов.
  • Qwen3-VL-4B–Bridge 29K: параметры и полное training state — optimizer, scheduler, Python/NumPy/PyTorch/CUDA RNG и порядок данных.
  • Жёсткое разделение common-start, continuation-training и evaluation seed; для Qwen точное продолжение требует оба 29K .pt.
  • Florence–Bridge: Residual-OPTR +500 даёт 59.38% overall против 42.19% у Uniform +500 на 384 роллаутах.

Пример

bash scripts/run_florence_bridge_arm.sh uniform uniform-reproduction
bash scripts/run_florence_bridge_arm.sh residual residual-reproduction \
  proposals/residual_temporal_proposal_k8_alpha0.2.json
ПодробнееБенчмарк: фиксированная no-video SimplerEnv WidowX/Bridge-панель, 4 policy seed × 4 задачи × 24 эпизода = 384 роллаута.
Железо и обучение: две GPU, BF16, ZeRO-2, micro-batch 4/GPU, accumulation 8, global batch 64, continuation seed 117465.
Ограничения: 20.5K model-only не хватает для точного продолжения — новые ветки только из полного 20K state. Для Qwen–Bridge нужны оба 29K .pt.
Идентичность эксперимента проверяется через loadability и semantic-проверки model/data/episode/anchor/support/batch/budget, без SHA-гейтов.
Discuss Telegram 1

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 17 GB · ready files19 GB
  • FP8 / INT8recommendedweights 8.6 GB9.9 GB
  • 4-bitweights 4.8 GB5.8 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 3080 · 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.

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

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

About the model

mmmmmcyaaa/ReTime
Author
mmmmmcyaaa
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments

Robotics

ACT Jenga2

ACT Jenga2