config.json

{
  "bottleneck": {
    "args": {
      "bottleneck_dim": 16,
      "norm": "none",
      "regularizer": {
        "args": {
          "codebook_loss_weight": 1.0,
          "codebook_size": 8192,
          "commitment_loss_weight": 0.25,
          "entropy_loss_temperature": 0.01,
          "entropy_loss_weight": 0.0,
          "l2_normalized": true,
          "stochastic": true,
          "stochastic_temperature": 0.03
        },
        "name": "vq"
      }
    },
    "name": "bottleneck"
  },
  "bottleneck_token_num": 1024,
  "decoder_depth": 12,
  "decoder_hidden_size": 768,
  "decoder_name": "none",
  "decoder_num_heads": 12,
  "decoder_patch_size": 8,
  "decoder_temporal_patch_size": 4,
  "encoder_depth": 12,
  "encoder_hidden_size": 768,
  "encoder_name": "none",
  "encoder_num_heads": 12,
  "encoder_query_gaussian_init": true,
  "frame_num": 16,
  "in_channels": 3,
  "input_size": 128,
  "latent_pe_scale_factor": 10000,
  "learned_decoder_latent_pe": false,
  "learned_decoder_patch_query_embed": false,
  "learned_encoder_latent_query_embed": true,
  "learned_encoder_patch_pe": false,
  "patch_size": 8,
  "prior_model": {
    "args": {
      "l2_normalized": true
    },
    "avg_loss_over_rounds": true,
    "latent_ce_temperature": 1.0,
    "mix_ss_max_ratio": 0.5,
    "mix_ss_peak_steps_ratio": 0.3,
    "n_rounds": 2,
    "name": "gptc-S",
    "no_dropout": false,
    "no_grad_before_last_round": false,
    "use_mix_ss": true
  },
  "query_init_std": 0.02,
  "temporal_patch_size": 4,
  "transformer_name": "transformer_encoder_parallel",
  "use_decoder_latent_token_type_embed": false,
  "use_decoder_patch_query_token_type_embed": true,
  "use_encoder_latent_query_token_type_embed": false,
  "use_encoder_patch_token_type_embed": false
}