Skip to content
Language models

Muse Shield er-v3a

LoRA для локального обезличивания данных

2 min read

Это PEFT-LoRA от TheEmperor2 для Muse Glimmer 30B — локальный движок приватности в шлюзе Muse Shield. Адаптер работает на вашей машине и превращает документ или сообщение в граф сущностей, чтобы облачная модель видела только плейсхолдеры вроде <PERSON_1>. Нужен, когда хочется пользоваться внешней LLM, не отправляя ей приватные данные.

Ключевые особенности

  • Находит не только PII: людей, организации, суммы, ID, контакты, адреса, даты, секреты, здоровье и др.
  • Строит entity graph: сущности, все упоминания (включая «он», «его счёт»), владельцев значений и связи.
  • Заменяет значения стабильными плейсхолдерами; соответствия хранятся в зашифрованном локальном vault.
  • По бенчмаркам 30B-модель локально близка к облачной Muse Spark по утечкам и обходит её по прямым идентификаторам, coreference и ownership.
  • Стандартный PEFT-адаптер: можно развернуть на Fireworks или self-hosted сервере с поддержкой LoRA для Muse Glimmer 30B.

Пример

hf download TheEmperor2/muse-shield-glimmer-er-v3a \
  --include 'adapter_config.json' 'adapter_model*' '*.json' 'chat_template.jinja' \
  --exclude 'results/*' --local-dir er-v3a
firectl create model muse-shield-glimmer-er-v3a-s250 ./er-v3a \
  --base-model accounts/fireworks/models/muse-glimmer-30b
Подробнее Nemotron-PII: core leak 2.0%, все типы утечек 9.5%; PUPA: 86.2% промптов полностью защищены.
TAB: прямые идентификаторы защищены на 98.9%, precision 85.1%; ER: coreference F1 0.974.
LoRA: rank 16, alpha 32, 12 160 обучающих строк; eval loss снизился с 0.1715 до 0.0037.
Ограничения: в основном английский текст; очень длинные документы иногда дают невалидный JSON, возможны пропуски повторных упоминаний.
Discuss Telegram 0 22

Will it run?

Working out the memory needs

This is a LoRA: it only runs on top of the base model meta/muse-glimmer-30b. The base model's size is still being collected; the estimate will appear within a day.

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.

    # TheEmperor2/muse-shield-glimmer-er-v3a
    # ===== 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 meta/muse-glimmer-30b --local-dir ~/models/muse-glimmer-30b
    hf download TheEmperor2/muse-shield-glimmer-er-v3a --local-dir ~/models/muse-shield-glimmer-er-v3a
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to muse-glimmer-30b
  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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

Requires the base model

meta/muse-glimmer-30b

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

New posts about versions of muse-glimmer-30b (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

TheEmperor2/muse-shield-glimmer-er-v3a
Author
TheEmperor2
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

220

Ask AI about this model

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

Comments