mirror of https://github.com/huggingface/diffusers.git synced 2026-01-29 07:22:12 +03:00

Files

dg845 b4220e97b1 Add Community Example Consistency Training Script (#6717 )

* initial commit for unconditional/class-conditional consistency training script

* make style

* Add entry for consistency training script in community README.

* Move consistency training script from community to research_projects/consistency_training

* Add requirements.txt and README to research_projects/consistency_training directory.

* Manually revert community README changes for consistency training.

* Fix path to script after moving script to research projects.

* Add option to load U-Net weights from pretrained model.

---------

Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

2024-01-26 15:10:57 +02:00

1.1 KiB

Raw Blame History

Consistency Training

train_cm_ct_unconditional.py trains a consistency model (CM) from scratch following the consistency training (CT) algorithm introduced in Consistency Models and refined in Improved Techniques for Training Consistency Models. Both unconditional and class-conditional training are supported.

A usage example is as follows:

accelerate launch examples/research_projects/consistency_training/train_cm_ct_unconditional.py \
    --dataset_name="cifar10" \
    --dataset_image_column_name="img" \
    --output_dir="/path/to/output/dir" \
    --mixed_precision=fp16 \
    --resolution=32 \
    --max_train_steps=1000 --max_train_samples=10000 \
    --dataloader_num_workers=8 \
    --noise_precond_type="cm" --input_precond_type="cm" \
    --train_batch_size=4 \
    --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
    --use_8bit_adam \
    --use_ema \
    --validation_steps=100 --eval_batch_size=4 \
    --checkpointing_steps=100 --checkpoints_total_limit=10 \
    --class_conditional --num_classes=10 \

1.1 KiB Raw Blame History

Consistency Training

1.1 KiB

Raw Blame History