Skip to content
Other

Gimbal-27B

открытая модель решений с калиброванными вероятностями

2 min read

Gimbal-27B от Superwheat — это 27B-модель (25,62 млрд параметров), дообученная через LoRA на Qwen/Qwen3.8-27B в роли 3-классного cross-encoder. Она принимает текстовый запрос с собственными вариантами и возвращает калиброванную вероятность для каждого варианта в задачах choice, noul (да/нет) и score. В репозитории лежат merged bf16-веса и исходный адаптер.

Ключевые особенности

  • Три типа типизированных решений: choice, noul и score — без свободного текста на выходе.
  • Калибровка: вероятность 0.8 соответствует примерно 80%; ECE на jev-dev choice — 1.9 с calibration v2.
  • Устойчивость к перестановке вариантов: 87.5 на JevBench против 76.4 у hosted Jev.
  • Готова к запуску через vLLM; есть merged bf16-checkpoint и оригинальный LoRA-адаптер.
  • На S1MB English (137 бенчмарков): 72.7 yes/no, 72.4 choice, 58.8 score против 64.6 / 67.2 / 46.9 у hosted Jev 1.13.
  • Лицензия Apache-2.0, как у базовой модели.

Пример

python3.12 -m venv .venv && source .venv/bin/activate
pip install huggingface-hub==1.33.0
hf download Superwheat/Gimbal-27B --local-dir gimbal-27b
pip install -r gimbal-27b/helper/requirements.txt
PYTHONPATH=gimbal-27b/helper python -m gimbal_serve.server --model gimbal-27b --port 8000
ПодробнееНа sealed held-out Gimbal-27B даёт 80.9% против 74.8% у hosted Jev 1.13 (+6.1).
JevBench consistency — 91.3 против 82.8; option-order stability — 87.5 против 76.4.
Нужны Linux, NVIDIA GPU с ~51 GB свободной VRAM и драйвер CUDA 13; сервер текстовый.
Слабое место — многошаговое применение формул: 50–62%; FP8-сборка тестировалась, но не опубликована.
Примеры
Discuss Telegram 0 112

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # Superwheat/Gimbal-27B
    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 Superwheat/Gimbal-27B --local-dir ~/models/Gimbal-27B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Gimbal-27B
  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 Gimbal-27B (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Superwheat/Gimbal-27B
Author
Superwheat
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1120

Ask AI about this model

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

Comments

Computer vision

Lucia v1

Lucia v1