stage_1:
  target: stage1.RAE
  params:
    encoder_cls: Dinov2withNorm
    encoder_config_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/huggingface/hub/models--facebook--dinov2-with-registers-base/snapshots/a1d738ccfa7ae170945f210395d99dde8adb1805
    encoder_input_size: 224
    encoder_params:
      dinov2_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/huggingface/hub/models--facebook--dinov2-with-registers-base/snapshots/a1d738ccfa7ae170945f210395d99dde8adb1805
      normalize: true
    decoder_config_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/LightningDiT/RAE/configs/decoder/ViTXL
    pretrained_decoder_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/LightningDiT/RAE/models/decoders/dinov2/wReg_base/ViTXL_n08/model.pt
    noise_tau: 0
    reshape_to_2d: true
    normalization_stat_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/LightningDiT/RAE/models/stats/dinov2/wReg_base/imagenet1k/stat.pt
stage_2:
  target: stage2.models.lightningDiT.LightningDiT
  params:
    input_size: 16
    patch_size: 1
    in_channels: 768
    hidden_size: 1152
    depth: 28
    num_heads: 16
    mlp_ratio: 4.0
    class_dropout_prob: 0.1
    num_classes: 1000
    use_qknorm: true
    use_swiglu: true
    use_rope: true
    use_rmsnorm: true
    wo_shift: false
    use_gembed: true
transport:
  params:
    path_type: Linear
    prediction: velocity
    loss_weight: null
    time_dist_type: logit-normal_0_1
sampler:
  mode: ODE
  params:
    sampling_method: euler
    num_steps: 50
    atol: 1.0e-06
    rtol: 0.001
    reverse: false
guidance:
  method: cfg
  scale: 1.0
  t_min: 0.0
  t_max: 1.0
misc:
  latent_size:
  - 768
  - 16
  - 16
  num_classes: 1000
  time_dist_shift_dim: 196608
  time_dist_shift_base: 4096
eval:
  eval_interval: 100000
  eval_model: true
  data_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/data/ILSVRC2012/ILSVRC2012_validation/data
  reference_npz_path: /home/jovyan/liushanyuan-sh-ceph/project/sub_project/gongyue/LightningDiT/guided-diffusion/evaluations/VIRTUAL_imagenet256_labeled.npz
training:
  epochs: 1400
  global_batch_size: 1024
  grad_accum_steps: 1
  ema_decay: 0.9995
  num_workers: 16
  log_interval: 100
  checkpoint_interval: 10
  sample_every: 10000
  clip_grad: 1.0
  global_seed: 42
  optimizer:
    lr: 0.0002
    betas:
    - 0.9
    - 0.95
    weight_decay: 0.0
  scheduler:
    type: linear
    warmup_epochs: 40
    decay_end_epoch: 800
    base_lr: 0.0002
    final_lr: 2.0e-05
    warmup_from_zero: false