{ "model_path": "adept_level1", "datatype": "adept", "dataset": "ADEPT", "learning_rate": 3.333e-05, "num_updates": 600000, "max_epochs": 1000, "max_updates": 800000, "phases": { "start_inner_loop": 60000, "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": 3, "error_dropout": 0.1 }, "bptt": { "bptt_start_timestep": 0, "bptt_steps": 2, "bptt_steps_max": 2, "increase_bptt_steps_every": 200000 }, "model": { "level": 1, "batch_size": 16, "num_objects": 7, "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": "standard" }, "decoder": { "channels": 48, "level1_channels": 24, "num_layers": 5 }, "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-06 } } }