Skip to content
Images

SDXL DPO Step 1000

UNet SDXL после diffusion DPO

1 min readupdated

Это чекпоинт step 1000: UNet от stabilityai/stable-diffusion-xl-base-1.0, дообученный diffusion DPO на доступном пуле HPDv3. В репозитории лежат FP16- и FP32-веса UNet, а также состояние чекпоинта и метаданные обучения.

Ключевые особенности

  • База — stabilityai/stable-diffusion-xl-base-1.0; обучение на 1 139 774 парах HPDv3, 150 пар с отсутствующими изображениями исключены.
  • 1024×1024, seed 42, effective pair batch size 2048, DPO beta 5000, peak learning rate 2.048e-5, Adafactor.
  • FP16-UNet в unet/ и FP32 master weights в unet_fp32/, плюс scaler.pt, random_states_*.pkl и training_metadata.json.
  • Это step-1000, не step-2000; optimizer.bin и scheduler.bin отсутствуют — полного optimizer-state для резюма нет.
  • Текстовые энкодеры, токенизатор и scheduler не дублируются: они загружаются из базовой SDXL.

Пример

import torch
from diffusers import AutoencoderKL, StableDiffusionXLPipeline, UNet2DConditionModel
repo = 'asksnskksnfapjgwanga/SDXL_DPO_Step_1000'
unet = UNet2DConditionModel.from_pretrained(repo, subfolder='unet', torch_dtype=torch.float16)
vae = AutoencoderKL.from_pretrained('madebyollin/sdxl-vae-fp16-fix', torch_dtype=torch.float16)
pipe = StableDiffusionXLPipeline.from_pretrained('stabilityai/stable-diffusion-xl-base-1.0', unet=unet, vae=vae, torch_dtype=torch.float16).to('cuda')
image = pipe('A watercolor painting of a quiet mountain village', height=1024, width=1024, num_inference_steps=50, guidance_scale=7.5, generator=torch.Generator(device='cuda').manual_seed(42)).images[0]
ПодробнееБенчмарков в карточке нет.
Обучение: 1 139 774 пары HPDv3, 1024×1024, DPO beta 5000, Adafactor, FP16 + FP32 master; случайный DDPM timestep на пару.
Ограничение: step-1000, optimizer.bin и scheduler.bin отсутствуют; для продолжения обучения оптимизатор нужно создавать заново.
VRAM не указана; для инференса нужны PyTorch, Diffusers, Transformers, Accelerate и Safetensors.
Discuss Telegram 0

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

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.

    # asksnskksnfapjgwanga/SDXL_DPO_Step_1000
    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 asksnskksnfapjgwanga/SDXL_DPO_Step_1000 --local-dir ~/models/SDXL_DPO_Step_1000
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to SDXL_DPO_Step_1000
  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.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • 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 (3)
    • 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.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • 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 SDXL_DPO_Step_1000 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

asksnskksnfapjgwanga/SDXL_DPO_Step_1000
Author
asksnskksnfapjgwanga
Category
Images
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