Skip to content
Other

10Eros Max H3 8-step TURBO beta7

int8 GPTQ для ComfyUI

1 min read

Это int8 GPTQ-квант версии 10Eros Max H3 8-step TURBO beta7 от tenstrip, опубликованный с разрешения автора. Формат — diffusion-single-file для ComfyUI: один safetensors на 22,5 GB вместо исходного bf16-чекпоинта на 40,2 GB.

Ключевые особенности

  • Один файл 10Eros_Max_h3_8stepTURBO_beta7_int8_gptq_activated.safetensors кладётся в ComfyUI/models/diffusion_models/.
  • GPTQ int8: квантованы 184 linear-слоя в блоках 2–47; блоки 0, 1, 48 и 49 остаются в bf16.
  • Hadamard-256 rotation, fp32 weight_scale на выходной канал, int8-веса и comfy_quant metadata для каждого слоя.
  • Калибровка на реальных активациях: 30 промптов примерно по 1 MP и 6 секунд; 24 использованы для подгонки, 6 — held-out только для оценки.
  • NRMSE ниже, чем у round-to-nearest int8: attn.qkv_proj примерно на 61%, mlp.fc1 на 60%, attn.out_proj на 57%, mlp.fc2 на 38%.

Пример

Скачайте safetensors и положите его в ComfyUI/models/diffusion_models/, затем выберите файл в ComfyUI как diffusion model. Отдельного кода запуска в карточке нет — это готовый single-file чекпоинт для папки моделей ComfyUI.

ПодробнееУменьшение ошибки указано per-layer по NRMSE относительно RTN int8; end-to-end бенчмарк качества не проводился.
Самый слабый отдельный слой — mlp.fc2 в блоке 30: снижение ошибки около 17%.
Исходная модель: tenstrip (10Eros Max H3). Лицензия — MiniMax-H3 Community License Agreement.
Discuss Telegram 2 837

Will it run?

Runs in INT8

Fits without quantisation.

  • INT8recommendedweights 21 GB · ready files24 GB
  • 4-bitweights 12 GB13 GB

Tip: go 8-bit.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: INT8 — from RTX 4090

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

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.

    # sandpies/10Eros-Max-H3-8stepTURBO-beta7-int8-GPTQ
    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 sandpies/10Eros-Max-H3-8stepTURBO-beta7-int8-GPTQ --local-dir ~/models/10Eros-Max-H3-8stepTURBO-beta7-int8-GPTQ
    # 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.

MiniMax H3 prompts

All prompts

Ready prompts with the resulting video: copy and try.

Model family

New posts about versions of 10Eros-Max-H3-8stepTURBO-beta7-int8-GPTQ (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

sandpies/10Eros-Max-H3-8stepTURBO-beta7-int8-GPTQ
Author
sandpies
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

8372

Ask AI about this model

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

Comments