BiliSakura / DiT-diffusers

BiliSakura/DiT-diffusers

Diffusers-ready checkpoints for Diffusion Transformers (DiT) , re-packaged for local/offline use with a project-owned custom DiTPipeline .

Re-distribution notice: weights and configs in this repo are re-distributed from facebook/DiT-XL-2-512 . Original work: Scalable Diffusion Models with Transformers (ICCV 2023) . License: CC BY-NC 4.0 .

This repo is derived from the development bundle in Visual-Generative-Foundation-Model-Collection . Inference only needs:

• This model repo ( BiliSakura/DiT-diffusers )
• PyPI diffusers , torch , safetensors

Important note

This repo intentionally does not use Diffusers built-in diffusers.DiTPipeline . Instead, each model subfolder contains pipeline.py with a custom class named DiTPipeline .

Available checkpoints

Each subfolder is a self-contained Diffusers model repo with:

• model_index.json (includes ImageNet id2label )
• pipeline.py (custom DiTPipeline )
• transformer/diffusion_pytorch_model.safetensors
• vae/diffusion_pytorch_model.safetensors
• scheduler/scheduler_config.json

Demo

from pathlib import Path
import torch
from diffusers import DiffusionPipeline

model_dir = Path("path/to/DiT-XL-2-512")
pipe = DiffusionPipeline.from_pretrained(
    str(model_dir),
    local_files_only=True,
    custom_pipeline=str(model_dir / "pipeline.py"),
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
).to("cuda")
generator = torch.Generator(device="cuda").manual_seed(0)
out = pipe(
    class_labels=[207],
    num_inference_steps=250,
    guidance_scale=4.0,
    generator=generator,
).images[0]
out

Repo layout

BiliSakura/DiT-diffusers/
├── README.md
├── DiT-XL-2-256/
└── DiT-XL-2-512/
    ├── README.md
    ├── model_index.json
    ├── pipeline.py
    ├── demo.png
    ├── transformer/
    │   ├── config.json
    │   └── diffusion_pytorch_model.safetensors
    ├── vae/
    │   ├── config.json
    │   └── diffusion_pytorch_model.safetensors
    └── scheduler/
        └── scheduler_config.json