# MAVT base model configuration
model:
  embed_dim: 768
  num_heads: 16
  num_blocks: 12
  patch_size: 16
  t_patch: 2
  latent_dim: 32
  kl_weight: 1.0e-4
  semantic_dim: 768
  dec_dim: 768
  num_dec_attn_blocks: 4
  r_s: 2
  r_t: 1
  use_gradient_checkpointing: true
  mlp_ratio: 4.0
  dropout: 0.0