Skip to content
Other

DINO-T2I SiT-XL

условная генерация по CLS-токену, не text-to-image

1 min readupdated

Финальный чекпоинт SiT-XL после 80 эпох обучения на ImageNet-1K в разрешении 256×256. Модель генерирует изображение, обусловленное замороженным CLS-представлением DINOv3 ViT-L/16 входной картинки; это не самостоятельный text-to-image pipeline.

Ключевые особенности

  • Условие — CLS-токен DINOv3 ViT-L/16, а не текст.
  • Архитектура SiT-XL: SwiGLU, RMSNorm, RoPE и shared adaLN.
  • Латентный автоэнкодер — FLUX.2 VAE; разрешение 256×256.
  • 80 эпох, optimizer step 200 160, batch size 512 на 4×NVIDIA H200, BF16.
  • В чекпоинте есть training- и EMA-модель, optimizer state, аргументы, эпоха и шаг.

Пример

Сценарий из карточки: исследователь кодирует входное изображение через DINOv3 ViT-L/16, берёт CLS-вектор и сэмплирует новое изображение 256×256. Для запуска нужны исходный код DINO-T2I, FLUX.2 VAE и энкодер DINOv3.

ПодробнееНа 25 held-out изображениях DINOv3-CLS similarity = 0.61471; это не ImageNet FID.
Чекпоинт весит ~7,3 ГБ; обучение шло с EMA decay 0.9995 и learning rate 2e-4.
Ограничение: не standalone text-to-image — нужны совместимый код, FLUX.2 VAE и DINOv3 ViT-L/16.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 6.8 GB · ready files8.0 GB
  • FP8 / INT8recommendedweights 3.4 GB4.3 GB
  • 4-bitweights 1.9 GB2.6 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.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 3080

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.

    # SaeedRazavi/dino-t2i-imagenet256-sitxl-flux2-dinocls-80ep
    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 SaeedRazavi/dino-t2i-imagenet256-sitxl-flux2-dinocls-80ep --local-dir ~/models/dino-t2i-imagenet256-sitxl-flux2-dinocls-80ep
    # 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 dino-t2i-imagenet256-sitxl-flux2-dinocls-80ep (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

SaeedRazavi/dino-t2i-imagenet256-sitxl-flux2-dinocls-80ep
Author
SaeedRazavi
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments