Skip to content
Other

Diorite-G4-E2B-v1.0

компактная модель решений на Gemma 4

2 min read

Diorite-G4-E2B-v1.0 — LoRA-адаптер от Leucodiorite на базе google/gemma-4-E2B-it. Это самая маленькая и быстрая модель в семействе Diorite v1.0: она получает контент и набор вопросов, за один forward pass выдаёт вероятность для каждого варианта ответа, а приложение превращает эти вероятности в решения. Адаптер содержит 94,9 млн параметров, весит 0,38 ГБ, базовые веса не меняются.

Ключевые особенности

  • Самая низкая задержка в семействе: медиана 36 мс на RTX PRO 6000.
  • LoRA rank 64, alpha 64 применяется к attention (q, v, o) и MLP-проекциям языковой модели; vision- и audio-модули не тронуты.
  • Выдаёт вероятности по всем вариантам каждого вопроса за один forward pass, а не генерирует свободный текст.
  • Адаптер на 94,9 млн параметров и 0,38 ГБ; минимум 24 ГБ GPU-памяти.
  • Работает через diorite-serve; в семействе также есть версии E4B, 12B и 31B.

Пример

Сценарий из карточки: сервис отправляет в diorite-serve контент и вопросы с вариантами ответов. Модель за один проход возвращает вероятности по каждому варианту, а приложение на их основе выбирает решение.

ПодробнееDecision Index 0.3: 43,17 против 25,91 у базовой gemma-4-E2B-it.
Accuracy — 64,0% против 53,9%, ECE — 0,077 против 0,362.
Минимум 24 ГБ VRAM; медиана 36 мс на RTX PRO 6000.
Это research preview для исследований и оценки на лидерборде Decision Index; лицензия адаптера — CC BY-NC 4.0.
Примеры
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 9.9 GB · ready files11 GB
  • FP8 / INT8recommendedweights 5.1 GB6.1 GB
  • 4-bitweights 3.0 GB3.8 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 5060 Ti 8GB · BF16 — from RTX 5070

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model gemma-4-E2B-it.

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.

    # Leucodiorite/Diorite-G4-E2B-v1.0
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 google/gemma-4-E2B-it --local-dir ~/models/gemma-4-E2B-it
    hf download Leucodiorite/Diorite-G4-E2B-v1.0 --local-dir ~/models/Diorite-G4-E2B-v1.0
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to gemma-4-E2B-it
  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

Requires the base model

google/gemma-4-E2B-it

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. gemma-4-E2Bfamily root
  2. gemma-4-E2B-it
  3. Diorite-G4-E2B-v1.0

New posts about versions of gemma-4-E2B (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

Leucodiorite/Diorite-G4-E2B-v1.0
Author
Leucodiorite
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