| { | |
| "seed": 42, | |
| "device": "cuda", | |
| "epochs": 5, | |
| "batch_size": 16, | |
| "lr": 3e-4, | |
| "kbbi_path": "kbbi_core_v2.json", | |
| "weight_decay": 1e-4, | |
| "grad_clip": 1.0, | |
| "log_every": 50, | |
| "eval_every": 200, | |
| "save_every": 1, | |
| "output_dir": "./aksara_output_train", | |
| "export_final": "./aksara_output_train/final_checkpoint.pt", | |
| "data_path": "data/corpus_state_train.jsonl", | |
| "val_path": "data/corpus_state_val.jsonl", | |
| "curriculum": { | |
| "phase_index": 0, | |
| "objectives": { | |
| "state_consistency": 1.0, | |
| "constraint_satisfaction": 1.0, | |
| "semantic_alignment": 1.0, | |
| "gos_coherence": 0.1, | |
| "multi_state_margin": 1.0 | |
| } | |
| }, | |
| "model": { | |
| "vocab_size": 5000, | |
| "affix_vocab_size": 40, | |
| "kbbi_input_dim": 16, | |
| "bsu_config": { | |
| "d_morpheme": 64, | |
| "d_semantic": 64, | |
| "d_role": 32, | |
| "d_context": 64, | |
| "dropout": 0.1 | |
| }, | |
| "meb_config": { | |
| "n_layers": 4, | |
| "n_dep_heads": 4, | |
| "dropout": 0.1 | |
| } | |
| } | |
| } | |