| { | |
| "model_type": "MultiModalHackVAE", | |
| "framework": "PyTorch", | |
| "task": "representation-learning", | |
| "dataset": "NetHack Learning Dataset", | |
| "latent_dim": 96, | |
| "lowrank_dim": 0, | |
| "architecture": "Multi-modal Variational Autoencoder for NetHack game states", | |
| "author": "Xu Chen", | |
| "description": "Advanced NetHack VAE", | |
| "tags": [ | |
| "nethack", | |
| "reinforcement-learning", | |
| "multimodal", | |
| "world-modeling", | |
| "vae" | |
| ], | |
| "use_cases": [ | |
| "Game state representation learning", | |
| "RL agent state abstraction", | |
| "NetHack gameplay analysis" | |
| ], | |
| "training_config": { | |
| "epochs": 15, | |
| "batch_size": 32, | |
| "max_learning_rate": 0.001, | |
| "sequence_size": 32, | |
| "shuffle_batches": true, | |
| "shuffle_within_batch": true, | |
| "vae_config": { | |
| "latent_dim": 96, | |
| "encoder_dropout": 0.1, | |
| "decoder_dropout": 0.1, | |
| "initial_mi_beta": 1.0, | |
| "final_mi_beta": 1.0, | |
| "mi_beta_shape": "constant", | |
| "initial_tc_beta": 0.0, | |
| "final_tc_beta": 10.0, | |
| "tc_beta_shape": "custom", | |
| "initial_dw_beta": 0.2, | |
| "final_dw_beta": 1.0, | |
| "dw_beta_shape": "custom", | |
| "warmup_epoch_ratio": 0.2, | |
| "free_bits": 0.75, | |
| "focal_loss_alpha": 0.5, | |
| "focal_loss_gamma": 2.0 | |
| }, | |
| "adaptive_weighting": { | |
| "initial_mi_beta": 1.0, | |
| "final_mi_beta": 1.0, | |
| "mi_beta_shape": "constant", | |
| "initial_tc_beta": 0.0, | |
| "final_tc_beta": 10.0, | |
| "tc_beta_shape": "custom", | |
| "initial_dw_beta": 0.2, | |
| "final_dw_beta": 1.0, | |
| "dw_beta_shape": "custom", | |
| "warmup_epoch_ratio": 0.2 | |
| }, | |
| "regularization": { | |
| "encoder_dropout": 0.1, | |
| "decoder_dropout": 0.1, | |
| "free_bits": 0.75, | |
| "focal_loss_alpha": 0.5, | |
| "focal_loss_gamma": 2.0 | |
| }, | |
| "early_stopping": { | |
| "enabled": false, | |
| "patience": 3, | |
| "min_delta": 0.01, | |
| "triggered": false, | |
| "best_epoch": null | |
| } | |
| }, | |
| "final_train_loss": 980.7457598876953, | |
| "final_test_loss": 990.1863067626953, | |
| "best_train_loss": 980.7457598876953, | |
| "best_test_loss": 985.553515625, | |
| "total_epochs": 15 | |
| } |