Skip to content
Other

Yoshino2 4B A2B

экспериментальная MoE-модель с reasoning-дистилляцией

1 min readupdated

Yoshino2 4B A2B — экспериментальная MoE-LLM от en5yme на базе MiniCPM5 1B. Всего 4 млрд параметров: четыре эксперта по 1B, два активны на токен. Модель создана для усиления reasoning у edge-моделей и дистиллирована в стиле NVIDIA Nemotron Lightning 30B A3B.

Ключевые особенности

  • Архитектура MoE: 4 эксперта MiniCPM5 1B, по два активных на токен; суммарно 4B параметров.
  • Reasoning-стиль дистиллирован из NVIDIA Nemotron Lightning 30B A3B.
  • Собрана через Mergekit-MoE из четырёх клонов MiniCPM5 1B.
  • Обучение: 10 817 синтетических Q/A-пар (около 135 МБ), Unsloth SFT, 3 эпохи QLoRA.
  • Модель помечена как highly experimental: без бенчмарков и не для production.
  • GGUF-квантизация через Llama.cpp может выдавать ошибки.

Пример

Карточка не приводит код запуска или пример промпта. Заявленный сценарий — reasoning-запросы для edge-моделей; качество нужно проверять вручную, так как бенчмарков нет.

Подробнее Бенчмарков на момент релиза нет — модель не позиционируется как production-grade.
Требования к VRAM/железу в карточке не указаны.
GGUF-квантизация через Llama.cpp может завершаться ошибками.
База — MiniCPM5 1B с downstream Apache 2.0; LICENSE в репозитории относится к файнтюну, не к базе.
Discuss Telegram 0 18

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.9 GB · ready files5.8 GB
  • FP8 / INT8recommendedweights 2.4 GB3.2 GB
  • 4-bitweights 1.4 GB2.0 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.

    # en5yme/Yoshino2-4B-A2B
    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 en5yme/Yoshino2-4B-A2B --local-dir ~/models/Yoshino2-4B-A2B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Yoshino2-4B-A2B
  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 MiniCPM5-1B (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

en5yme/Yoshino2-4B-A2B
Author
en5yme
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

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

Comments