RSEdit-UNet-token-concat

RSEdit UNet Concat Ablation

This directory stores checkpoints for the RSEdit UNet concatenation ablation run.

The active/exported checkpoint is under:

• token_concat/

token_concat/ contents

• model_index.json: pipeline/component registry (RSEditUNetTokenConcatPipeline).
• pipeline.py: custom local pipeline implementation used for inference/loading.
• checkpoint-30000/: training-state snapshot at step 30k (optimizer/scheduler/random state plus UNet weights).
• unet/, vae/, text_encoder/, tokenizer/, scheduler/: exported model components for direct pipeline loading.
• feature_extractor/, safety_checker/: SD-style auxiliary components referenced by this pipeline.
• logs/: experiment logs (token_concat/logs/...).

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 mainly needed for resume training rather than inference.