Skip to content
Other

Gemma-4-26B-A4B MAPLE

сжатый MoE для кода на 6 ГБ VRAM

1 min read

Это сжатый GGUF-контейнер Gemma 4 26B-A4B от AmSach, построенный методом MAPLE (Manifold-Activated Pruning of Latent Experts). Модель заточена под Python-программирование, синтез алгоритмов, структуры данных и формальные STEM-рассуждения; исходные 9.92 ГБ сжаты до 4.75 ГБ.

Ключевые особенности

  • Специализация: Python, синтез алгоритмов, структуры данных и формальные STEM-рассуждения.
  • MAPLE удаляет 80 «спящих» экспертов из 128 на каждом из 30 слоёв, оставляя 48 экспертов и 79.88% активной массы forward-pass.
  • Контейнер 4.75 ГБ (IQ2_XXS): 10.95B физических параметров при 3.47 BPW и 1.48 эффективных бита на параметр относительно 25.6B базы.
  • 100% слоёв в VRAM на RTX 4050 6 ГБ: 45.4 tok/s и 22.0 мс/токен.
  • Пайплайн domain-agnostic: профилирование через llama-imatrix и сборка контейнера под свой корпус.

Пример

llama-cli -hf AmSach/Gemma-4-26B-A4B-MAPLE-Sub5GB-GGUF -ngl 99 -c 2048 -n 512 --temp 0.4 --top-p 0.9 -r "<turn|>" -p "<|turn>user\nExplain Rayleigh scattering.\n<turn|>\n<|turn>model\n"
ПодробнееБенчмарк на RTX 4050 6GB: 45.4 tok/s против 6.9 у базовой 128-expert (6.57x), 4.75 ГБ вместо 9.92 ГБ.
В VRAM: 4,820 MiB при контексте 2048, остаётся ~1.29 GiB запаса.
Сжатие: 48 из 128 экспертов на слой, 79.88% активной массы forward-pass.
Ограничение: заточен под Python/алгоритмы/STEM; мультиязычные и общие диалоговые эксперты удалены.
Discuss Telegram 2 213

Will it run?

Runs in GGUF IQ2_XXS

The most precise of the ready quantised files fits.

  • GGUF IQ2_XXSweights 4.4 GB · ready files5.4 GB

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF IQ2_XXS — 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.

    # AmSach/Gemma-4-26B-A4B-MAPLE-Sub5GB-GGUF
    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 AmSach/Gemma-4-26B-A4B-MAPLE-Sub5GB-GGUF --local-dir ~/models/Gemma-4-26B-A4B-MAPLE-Sub5GB-GGUF
    # 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

  1. gemma-4-26B-A4Bfamily root
  2. gemma-4-26B-A4B-it
  3. Gemma-4-26B-A4B-MAPLE-Sub5GB-GGUF

New posts about versions of gemma-4-26B-A4B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
Merges 5All on Hugging Face

About the model

AmSach/Gemma-4-26B-A4B-MAPLE-Sub5GB-GGUF
Author
AmSach
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

2132

Ask AI about this model

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

Comments