Skip to content
Other

Kimi K3.1-Pro

abliteration-патч для снижения отказов

1 min readupdated

pqhaz выпустил экспериментальный весовой патч для закреплённой ревизии moonshotai/Kimi-K3.1-Pro. Это не самостоятельная модель: результат получается после применения точного сжатого патча к исходному checkpoint и загрузки в совместимый inference runtime Kimi K3.1-Pro.

Ключевые особенности

  • Патч применяет постоянные проекции к 278 BF16-матрицам attention/output-expert; квантованные routed expert tensors, embeddings, имена тензоров и заголовки checkpoint сохранены.
  • Выбран вариант trace4: это частичное снижение отказов, а не заявление об их полном устранении или сохранении всех способностей модели.
  • В измерении отказы снизились с 86/89 до 37/89 после «запекания», содержательные ответы выросли с 3/89 до 52/89.
  • Benign completions остались 58/60, HumanEval — 30/30; проверяются хэши индекса, исходных, сжатых и восстановленных тензоров.
  • Для использования результата не нужен prompt или reasoning prefill.

Пример

python -m pip install -r requirements.txt
hf download moonshotai/Kimi-K3.1-Pro --revision 6d0a0202956796543364d666ab19c6ad1a494ee1 --local-dir /data/k31-base
python apply_exact_patch.py --base /data/k31-base --patch ./patch --out /data/k31-edited
ПодробнееСжатая дельта — 14.70 ГБ; для результата нужно около 1.56 ТБ под независимый checkpoint.
Тесты: 8×MI355X, ROCm vLLM-образ, low reasoning effort, контекст 4096 токенов; другое железо и настройки не валидировались.
Оценка внутренняя: 89 запросов, один LLM-судья, малые выборки; это не публичный бенчмарк.
Discuss Telegram 2

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 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.

    # pqhaz/Kimi-K3.1-Pro-abliteration-patch-20261105
    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 pqhaz/Kimi-K3.1-Pro-abliteration-patch-20261105 --local-dir ~/models/Kimi-K3.1-Pro-abliteration-patch-20261105
    # 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 Kimi-K3.1-Pro (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face

About the model

pqhaz/Kimi-K3.1-Pro-abliteration-patch-20261105
Author
pqhaz
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

02

Ask AI about this model

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

Comments