{
  "model_path": "clevrer_level1",
  "datatype": "clevrer",
  "dataset": "CLEVRER",
  "learning_rate": 3.3333e-05,
  "num_updates": 200000,
  "max_epochs": 1000,
  "max_updates": 600000,
  "phases": {
     "start_inner_loop": 100000,
     "shufleslots_end": 30000,
     "entity_pretraining_phase2_end": 50000,
     "entity_pretraining_phase1_end": 30000,
     "background_pretraining_end": 0
  },
  "defaults": {
     "num_workers": 2,
     "prefetch_factor": 2,
     "statistics_offset": 10,
     "load_optimizers": false,
     "teacher_forcing": 10,
     "skip_frames": 2,
     "error_dropout": 0.1
  },
  "bptt": {
     "bptt_start_timestep": 0,
     "bptt_steps": 3,
     "bptt_steps_max": 3,
     "increase_bptt_steps_every": 200000
  },
  "model": {
   "level": 1,
   "batch_size": 32,
   "num_objects": 6,
   "img_channels": 3,
   "input_size": [
    320,
    480
   ],
   "latent_size": [
    20,
    30
   ],
   "gestalt_size": 96,
   "bottleneck": "binar",
   "position_regularizer": 0.01,
   "time_regularizer": 0.1,
   "encoder_regularizer": 0.333333,
   "inner_loop_enabled": true,
   "encoder": {
    "channels": 48,
    "level1_channels": 24,
    "num_layers": 3,
    "reg_lambda": 1e-10
   },
   "predictor": {
    "heads": 2,
    "layers": 2,
    "channels_multiplier": 2,
    "reg_lambda": 1e-10,
    "transformer_type": "shared"
   },
   "decoder": {
    "channels": 48,
    "level1_channels": 24,
    "num_layers": 3
   },
   "background": {
    "learning_rate": 0.0001,
    "learning_rate_old": 0.0001,
    "num_layers": 1,
    "latent_channels": 48,
    "level1_channels": 24,
    "gestalt_size": 8,
    "flow": false
   },
   "update_module": {
    "reg_lambda": 5e-08  
   }
  }
 }