model:
  target: Models.interpretable_diffusion.gaussian_diffusion.Diffusion_TS
  params:
    seq_length: 100
    feature_size: 14
    n_layer_enc: 3
    n_layer_dec: 3
    d_model: 64  # 4 X 16
    timesteps: 500
    sampling_timesteps: 500
    loss_type: 'l1'
    beta_schedule: 'cosine'
    n_heads: 4
    mlp_hidden_times: 4
    attn_pd: 0
    resid_pd: 0
    kernel_size: 1
    padding_size: 0

solver:
  base_lr: 1.0e-5
  max_epochs: 12000
  results_folder: ./Checkpoints_mujoco_sssd
  gradient_accumulate_every: 2
  save_cycle: 1200  # max_epochs // 10
  ema:
    decay: 0.995
    update_interval: 10
  
  scheduler:
    target: engine.lr_sch.ReduceLROnPlateauWithWarmup
    params:
      factor: 0.5
      patience: 3000
      min_lr: 1.0e-5
      threshold: 1.0e-1
      threshold_mode: rel
      warmup_lr: 8.0e-4
      warmup: 500 
      verbose: False