File size: 958 Bytes
be9fa39 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 |
your_local_path='output'
hf download Efficient-Large-Model/SANA_Sprint_1.6B_1024px_teacher_diffusers --local-dir $your_local_path/SANA_Sprint_1.6B_1024px_teacher_diffusers
# or Sana_Sprint_0.6B_1024px_teacher_diffusers
python train_sana_sprint_diffusers.py \
--pretrained_model_name_or_path=$your_local_path/SANA_Sprint_1.6B_1024px_teacher_diffusers \
--output_dir=$your_local_path \
--mixed_precision=bf16 \
--resolution=1024 \
--learning_rate=1e-6 \
--max_train_steps=30000 \
--dataloader_num_workers=8 \
--dataset_name='brivangl/midjourney-v6-llava' \
--file_path data/train_000.parquet data/train_001.parquet data/train_002.parquet \
--checkpointing_steps=500 --checkpoints_total_limit=10 \
--train_batch_size=1 \
--gradient_accumulation_steps=1 \
--seed=453645634 \
--train_largest_timestep \
--misaligned_pairs_D \
--gradient_checkpointing \
--resume_from_checkpoint="latest" \
|