Skip to content
Computer vision

Muse-Glimmer-30B-uncensored2

30B-агент с минимумом отказов

2 min readupdated

Это decensored-версия Muse-Glimmer-30B от electroglyph, полученная через Heretic v2.2.0.dev0. База — плотный causal transformer на ~29,6B параметров с perception-энкодером на ~1,8B: вход текст + изображения, выход текст, контекст 131 072+ токена. Модель заточена под автономных агентов, tool use и локальный запуск.

Ключевые особенности

  • Снижены отказы: 10/100 против 98/100 у оригинала; KL divergence — 0,0425.
  • Мультимодальный вход: текст и изображения через perception-энкодер, до 4096 visual tokens на изображение.
  • Агентные сценарии: tool use, multi-step reasoning, восстановление после ошибок инструментов, совместимость с OpenClaw и Hermes Agent.
  • Локальный запуск: 4-bit K-Quant-17GB занимает около 17 ГБ и рассчитан на 24/32 ГБ VRAM; speculative decoding DFlash ускоряет генерацию.
  • Управляемая сила рассуждений: low / medium / high / xhigh через system prompt.

Пример

Reasoning strength: high
Попросите агента разобрать скриншот интерфейса, найти ошибку и вызвать инструмент для исправления. Модель должна считать изображение, построить план, выполнить вызов и при сбое повторить попытку с диагностикой.

ПодробнееОтказы: 10/100 против 98/100 у базовой модели; KL divergence — 0,0425.
4-bit K-Quant-17GB: около 17 ГБ, до 1% деградации по 15 бенчмаркам; целевое железо — 24 ГБ VRAM.
Контекст 131 072+ токена, knowledge cutoff — 4 января 2026.
DFlash-драфтер даёт до 3,1× на RTX 5090 и 1,5–1,8× на Apple M4/M5 Max.
Это decensored-вариант: для продакшена карточка базовой модели рекомендует внешние guardrails.
Discuss Telegram 1

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 55 GB · ready files62 GB
  • FP8 / INT8recommendedweights 28 GB31 GB
  • 4-bitweights 16 GB18 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.

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.

    # electroglyph/Muse-Glimmer-30B-uncensored2
    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 electroglyph/Muse-Glimmer-30B-uncensored2 --local-dir ~/models/Muse-Glimmer-30B-uncensored2
    # 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

New posts about versions of Muse-Glimmer-30B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face
Merges 1All on Hugging Face

About the model

electroglyph/Muse-Glimmer-30B-uncensored2
Author
electroglyph
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments