27 lines
937 B
Bash
27 lines
937 B
Bash
export WANDB_KEY=""
|
|
export ENTITY=""
|
|
export PROJECT="t2v-f16s3-128-imgvae188-bf16-ckpt-xformers"
|
|
accelerate launch \
|
|
--config_file scripts/accelerate_configs/ddp_config.yaml \
|
|
opensora/train/train_t2v.py \
|
|
--model LatteT2V-XL/122 \
|
|
--text_encoder_name DeepFloyd/t5-v1_1-xxl \
|
|
--dataset t2v \
|
|
--ae stabilityai/sd-vae-ft-mse \
|
|
--data_path /root/autodl-tmp/sea.csv \
|
|
--video_folder /root/autodl-tmp/sea \
|
|
--sample_rate 3 \
|
|
--num_frames 16 \
|
|
--max_image_size 128 \
|
|
--gradient_checkpointing \
|
|
--attention_mode xformers \
|
|
--train_batch_size=8 --dataloader_num_workers 10 \
|
|
--gradient_accumulation_steps=1 \
|
|
--max_train_steps=1000000 \
|
|
--learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
|
|
--mixed_precision="bf16" \
|
|
--report_to="wandb" \
|
|
--checkpointing_steps=500 \
|
|
--output_dir="t2v-f16s3-128-imgvae188-bf16-ckpt-xformers" \
|
|
--allow_tf32
|