| { | |
| "method": "ewc", | |
| "seed": 20260907, | |
| "order": [ | |
| "T1", | |
| "T3", | |
| "T2", | |
| "T4" | |
| ], | |
| "legacy_epochs_per_task": 30, | |
| "requested_samples_per_epoch": 4194304, | |
| "actual_generated_samples_per_epoch": 2097152, | |
| "steps_per_task": 15360, | |
| "micro_batch_size": 2048, | |
| "accumulate_steps": 2, | |
| "effective_batch_size": 4096, | |
| "total_optimizer_steps": 61440, | |
| "samples_processed": 251658240, | |
| "validation_samples_per_stage": 65536, | |
| "method_parameters": { | |
| "lambda": 316.22776601683796, | |
| "gamma": 1.0, | |
| "positive_weight": 1.0, | |
| "fisher_estimator": "mean_squared_per_sample_bce_gradient", | |
| "fisher_samples": 4096, | |
| "fisher_batch_size": 2048, | |
| "fisher_per_sample_chunk_size": 16, | |
| "fisher_seed": 20260912, | |
| "online_state_bytes": 5246466 | |
| }, | |
| "initial_checkpoint": "baselines/initial-c0/model.safetensors", | |
| "training_distance": 9, | |
| "training_rounds": 9, | |
| "no_ood_or_willow_used": true | |
| } | |