configs/opensora/train/16x256x256-mask.py

# Define dataset
dataset = dict(
    type="VideoTextDataset",
    data_path=None,
    num_frames=16,
    frame_interval=3,
    image_size=(256, 256),
)

# Define acceleration
num_workers = 4
dtype = "bf16"
grad_checkpoint = True
plugin = "zero2"
sp_size = 1

# Define model
model = dict(
    type="STDiT-XL/2",
    space_scale=0.5,
    time_scale=1.0,
    from_pretrained="PixArt-XL-2-512x512.pth",
    enable_flash_attn=True,
    enable_layernorm_kernel=True,
)
mask_ratios = {
    "identity": 0.7,
    "random": 0.15,
    "mask_head": 0.05,
    "mask_tail": 0.05,
    "mask_head_tail": 0.05,
}
vae = dict(
    type="VideoAutoencoderKL",
    from_pretrained="stabilityai/sd-vae-ft-ema",
)
text_encoder = dict(
    type="t5",
    from_pretrained="DeepFloyd/t5-v1_1-xxl",
    model_max_length=120,
    shardformer=True,
)
scheduler = dict(
    type="iddpm",
    timestep_respacing="",
)

# Others
seed = 42
outputs = "outputs"
wandb = False

epochs = 1000
log_every = 10
ckpt_every = 1000
load = None

batch_size = 8
lr = 2e-5
grad_clip = 1.0