Skip to content
Other

Arbora: Qwen3-8B online expert

экспертные веса для спекулятивного декодирования

1 min readupdated

Это часть кодового релиза статьи Arbora о параллелизме block diffusion drafters в speculative decoding (статья на анонимном ревью). Чекпоинт содержит последние версии двух online-learning прогонов на потоке SWE-запросов для Qwen3-8B: добавленный Arbora expert, обученный вместе с path scorer, и вариант только с новым expert — drafter и остальные эксперты заморожены.

Ключевые особенности

  • Не самостоятельная LLM: загружается в serving-стек qwen3-8b-online поверх arbora-qwen3-8b-drafter и arbora-qwen3-8b-experts.
  • Два варианта: new-expert-and-path-scorer с обученным selector и new-expert только с новым expert.
  • В составе: experts.safetensors и selector.safetensors для первого варианта; во втором path scorer остаётся arbora-qwen3-8b-path-scorer.
  • Обучение шло на ответах целевой модели на закрытом наборе промптов; рецепт и скрипты пересборки данных описаны в кодовом релизе.
  • Опубликованных бенчмарков и метрик в карточке нет — только артефакты для online-стека.

Пример

Сценарий из карточки: в eval/online/ веса подгружаются на месте в стек qwen3-8b-online. Для new-expert-and-path-scorer берут experts.safetensors и selector.safetensors, для new-expert — только experts.safetensors, а path scorer остаётся неизменным.

ПодробнееБенчмарков в карточке нет.
Данные обучения: ответы целевой модели на неопубликованный набор промптов; код релиза описывает рецепт и скрипты пересборки.
Лицензия Apache 2.0, сторонние уведомления в NOTICE.
Нужен совместимый serving-стек qwen3-8b-online; отдельные требования к VRAM не указаны.
Discuss Telegram 0

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.

    # arbora-anon/arbora-qwen3-8b-online-expert
    # ===== 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 arbora-anon/arbora-qwen3-8b-online-expert --local-dir ~/models/arbora-qwen3-8b-online-expert
    # 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 arbora-qwen3-8b-online-expert (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

arbora-anon/arbora-qwen3-8b-online-expert
Author
arbora-anon
Category
Other
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