BiliSakura/RSEdit-DiT-channel-concat

RSEdit DiT Concat Ablation (channel_concat)

This directory stores a local Diffusers-compatible checkpoint for the RSEdit DiT channel-concatenation ablation run.

Contents

• model_index.json: pipeline/component registry (RSEditDiTChannelConcatPipeline).
• pipeline.py: custom local pipeline implementation used for inference/loading.
• checkpoint-30000/: training-state snapshot at step 30k (includes optimizer/scheduler/random states and transformer checkpoint).
• transformer/, vae/, text_encoder/, tokenizer/, scheduler/: exported model components for direct pipeline loading.
• logs/: experiment logs (logs/rsedit-dit/...).

Quick Load (Diffusers)

from diffusers import DiffusionPipeline

model_dir = "path/to/model"
pipe = DiffusionPipeline.from_pretrained(model_dir, trust_remote_code=True)
pipe = pipe.to("cuda")

Notes

• checkpoint-30000/optimizer.bin is large and is only needed to resume training, not for inference.
• Keep pipeline.py and model_index.json in sync if the custom pipeline class name changes.