Skip to content
Computer vision

Qwen3.8-27B-Quickthinker

думает на 43% короче без потери точности

1 min read

Мультимодальная LLM на базе Qwen/Qwen3.8-27B с небольшим изменением весов: модель тратит заметно меньше токенов на рассуждения при той же accuracy. Это не обучение и не LoRA — патч весов, который можно применить к собственной копии базовой модели.

Ключевые особенности

  • Средняя длина thinking снижена с 1 951 до 1 116 токенов (−42,8%) на 450 отложенных задачах.
  • Accuracy осталась на уровне: 0,913 → 0,928; авторы просят читать это как «та же», а не «лучше».
  • Меньше «убежавших» трасс: 1,6% прогонов упираются в лимит 16k токенов против 3,8% у оригинала.
  • Не нужен новый system prompt: эффект зашит в веса, chat template и вызов не меняются.
  • Совместимость с Qwen3.8-27B: та же архитектура, tokenizer, vision tower и Apache-2.0.

Пример

vllm serve scar-ai/Qwen3.8-27B-Quickthinker --reasoning-parser qwen3 --max-model-len 32768
ПодробнееСредняя длина thinking: 1 951 → 1 116 токенов (−42,8%); accuracy 0,913 → 0,928, но без учёта обрезанных прогонов разница в пределах шума.
По задачам: GSM8K — −69% thinking, MATH-500 — −43%, MMLU-Pro — −38%.
Веса ~54 ГБ в bf16: нужен GPU класса 80 ГБ; запуск как у Qwen3.8-27B.
Обучения не было: изменены 71 680 из 26,9 млрд весов (0,00027%) через weight-space programming и miniml.
Discuss Telegram 0

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 52 GB · ready files57 GB
  • FP8 / INT8recommendedweights 26 GB29 GB
  • 4-bitweights 15 GB17 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.

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.

    # scar-ai/Qwen3.8-27B-Quickthinker
    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 scar-ai/Qwen3.8-27B-Quickthinker --local-dir ~/models/Qwen3.8-27B-Quickthinker
    # 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 Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

scar-ai/Qwen3.8-27B-Quickthinker
Author
scar-ai
Category
Computer vision
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