Skip to content
Other

Kimi-K3.1-Pro-Coder

2.8T MoE в смешанных битах для кода и математики

2 min read

Это полная 96-шардовая квантизация moonshotai/Kimi-K3.1-Pro от Grawnch: 93 слоя, 896 routed experts на слой MoE, top_k=16 LatentMoE с SiTU-GLU, KDA linear attention + gated NoPE MLA и AttnRes. Все non-routed веса оставлены в BF16/F32, а 82 432 routed эксперта сжаты по трёхуровневой схеме Strata-Coder, калиброванной на коде и математике.

Ключевые особенности

  • Три уровня экспертов: hot coding/math, warm general и cold tail — от 2.81 до 1.44 bpw.
  • Полный размер — 774.97 GB / 721.75 GiB, на 56.2% меньше базовой модели.
  • Non-routed часть, включая attention, shared experts, lm_head и vision tower, сохранена в BF16/F32.
  • Заявлена работа с 3-уровневым offloading: VRAM + RAM + NVMe через загрузчик Strata и ядра dr-triton.
  • Калибровка и оценка ориентированы на coding/math-задачи, а не на общий чат.

Пример

В карточке нет готового сниппета запуска. Сценарий из оценки: модель решает задачи AIME 2025 с ответом в \boxed{} и проходит LiveCodeBench-задачи с проверкой unit-тестов в режиме Pass@1.

ПодробнееНа NVIDIA RTX PRO 6000 Blackwell (102.6 GB GDDR7 VRAM): AIME 2025 — 29/30, LiveCodeBench — 29/30, Top-5 agreement — 99.91–99.92%.
Средний KLD — около 0.020 nats, полный 93-слойный батч — 36.77 tok/s.
Запуск описан через 3-уровневую схему VRAM + RAM + NVMe, загрузчик Strata и fused Triton-ядра dr-triton.
Discuss Telegram 0

Will it run?

Won't run

Not enough memory even for 4-bit. Below are GPUs you can rent.

  • U8weights 1250 GB · ready files1376 GB
  • FP8 / INT8recommendedweights 1183 GB1302 GB
  • 4-bitweights 665 GB732 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

No single instance holds the whole model: it needs RAM offload or a cluster.

Your own GPU: Won't fit any home GPU — renting is cheaper.

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.

    # Grawnch/Kimi-K3.1-Pro-Coder
    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 Grawnch/Kimi-K3.1-Pro-Coder --local-dir ~/models/Kimi-K3.1-Pro-Coder
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Kimi-K3.1-Pro-Coder
  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

Grawnch/Kimi-K3.1-Pro-Coder
Author
Grawnch
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