43 lines
1.5 KiB
Bash
43 lines
1.5 KiB
Bash
export MODEL_NAME="models/Diffusion_Transformer/PixArt-XL-2-512x512"
|
|
export DATASET_NAME="datasets/internal_datasets/"
|
|
export DATASET_META_NAME="datasets/internal_datasets/metadata.json"
|
|
export NCCL_IB_DISABLE=1
|
|
export NCCL_P2P_DISABLE=1
|
|
NCCL_DEBUG=INFO
|
|
|
|
# When train model with multi machines, use "--config_file accelerate.yaml" instead of "--mixed_precision='bf16'".
|
|
# vae_mode can be choosen in "normal" and "magvit"
|
|
# transformer_mode can be choosen in "normal" and "kvcompress"
|
|
accelerate launch --mixed_precision="bf16" scripts/train_t2iv.py \
|
|
--pretrained_model_name_or_path=$MODEL_NAME \
|
|
--train_data_dir=$DATASET_NAME \
|
|
--train_data_meta=$DATASET_META_NAME \
|
|
--config_path "config/easyanimate_video_motion_module_v1.yaml" \
|
|
--image_sample_size=512 \
|
|
--video_sample_size=512 \
|
|
--video_sample_stride=2 \
|
|
--video_sample_n_frames=16 \
|
|
--train_batch_size=2 \
|
|
--video_repeat=1 \
|
|
--image_repeat_in_forward=4 \
|
|
--gradient_accumulation_steps=1 \
|
|
--dataloader_num_workers=8 \
|
|
--num_train_epochs=100 \
|
|
--checkpointing_steps=500 \
|
|
--validation_prompts="A girl with delicate face is smiling." \
|
|
--validation_epochs=1 \
|
|
--validation_steps=100 \
|
|
--learning_rate=2e-05 \
|
|
--lr_scheduler="constant_with_warmup" \
|
|
--lr_warmup_steps=100 \
|
|
--seed=42 \
|
|
--output_dir="output_dir" \
|
|
--enable_xformers_memory_efficient_attention \
|
|
--gradient_checkpointing \
|
|
--mixed_precision="bf16" \
|
|
--adam_weight_decay=3e-2 \
|
|
--adam_epsilon=1e-10 \
|
|
--max_grad_norm=1 \
|
|
--vae_mini_batch=16 \
|
|
--use_ema \
|
|
--trainable_modules "attn_temporal" |