Skip to content
Other

GLM-4.7-Flash-abliterated

30B MoE с ослабленными отказами

1 min readupdated

Это abliterated-версия zai-org/GLM-4.7-Flash от Torbeck: 30B параметров, 3B активных на токен, архитектура GLM-4 MoE Lite с Multi-head Latent Attention (MLA) и контекстом 128K. Модель сделана через Abliterix, чтобы снизить число отказов, и заявлена для исследований.

Ключевые особенности

  • Архитектура: GLM-4 MoE Lite, 47 слоёв (1 dense + 46 MoE), 64 routed + 1 shared expert, top-4 routing.
  • Контекст: 128K токенов; точность BF16.
  • Отказы: 1/100 против 92/100 у оригинала по оценке LLM-судьи на 100 вредных промптах.
  • KL divergence: 0.0133 — общие способности почти как у исходной модели.
  • Запуск: BF16 ~56 GB VRAM, INT8 ~30 GB, NF4 ~15 GB.

Пример

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "Torbeck/GLM-4.7-Flash-abliterated",
    torch_dtype="auto", device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("Torbeck/GLM-4.7-Flash-abliterated", trust_remote_code=True)
ПодробнееKL divergence 0.0133; отказы 1/100 против 92/100 у оригинала.
VRAM: BF16 ~56 GB (A100 80GB/H100), INT8 ~30 GB (A40/RTX 4090), NF4 ~15 GB (RTX 3090/4080).
Лицензия upstream — MIT; это независимый производный чекпойнт, не официальный релиз.
Только для исследований: защитные ограничения ослаблены или сняты, возможен небезопасный контент.
Discuss Telegram 1 14

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 56 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.

    # Torbeck/GLM-4.7-Flash-abliterated
    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 Torbeck/GLM-4.7-Flash-abliterated --local-dir ~/models/GLM-4.7-Flash-abliterated
    # 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 GLM-4.7-Flash (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Torbeck/GLM-4.7-Flash-abliterated
Author
Torbeck
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

141

Ask AI about this model

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

Comments