| { |
| "phase_results": [ |
| { |
| "phase": "simple_multiturn_transcript", |
| "best_selection_metric_value": 6.045462267592054, |
| "best_metrics": { |
| "validation/loss_total": 1.7929038028086592, |
| "validation/loss_response": 2.106443265529265, |
| "validation/loss_current_user_reconstruction": 1.6467093680787979, |
| "validation/loss_probe": 1.3120193120973094, |
| "validation/response_first_token_exact_match": 0.1518987341772152, |
| "validation/current-user_reconstruction_first_token_exact_match": 0.22784810126582278, |
| "validation/probe_first_token_exact_match": 1.0, |
| "validation/supervised_turn_count": 474.0, |
| "validation/response_exact_match": 0.007936507936507936, |
| "validation/response_similarity": 0.0982966461414312, |
| "validation/response_token_f1": 0.18865678434635447, |
| "validation/response_line_recall": 0.007936507936507936, |
| "validation/response_reconstruction_similarity": 0.2672819842355101, |
| "validation/response_reconstruction_exact_match": 0.0, |
| "validation/response_reconstruction_token_f1": 0.2672819842355101, |
| "validation/probe_exact_match": 0.0, |
| "validation/probe_transcript_similarity": 0.23611152152229414, |
| "validation/probe_token_f1": 0.4722230430445883, |
| "validation/probe_line_recall": 0.0, |
| "validation/response_similarity_by_turn/turn_2": 0.1278913959770529, |
| "validation/response_similarity_by_turn/turn_3": 0.09627094558882597, |
| "validation/response_similarity_by_turn/turn_4": 0.07133303521592968, |
| "validation/response_similarity_by_turn/turn_5": 0.08668160657233656, |
| "validation/response_similarity_by_turn/turn_6": 0.060287911253460294, |
| "validation/response_reconstruction_similarity_by_turn/turn_2": 0.4131726085584623, |
| "validation/response_reconstruction_similarity_by_turn/turn_3": 0.2114348366719477, |
| "validation/response_reconstruction_similarity_by_turn/turn_4": 0.20395226248593826, |
| "validation/response_reconstruction_similarity_by_turn/turn_5": 0.16776266235940282, |
| "validation/response_reconstruction_similarity_by_turn/turn_6": 0.1779640527539687, |
| "validation/probe_transcript_similarity_by_turn/turn_2": 0.2543098189406736, |
| "validation/probe_transcript_similarity_by_turn/turn_3": 0.22976204949484447, |
| "validation/probe_transcript_similarity_by_turn/turn_4": 0.22848382794491162, |
| "validation/probe_transcript_similarity_by_turn/turn_5": 0.22902170973243605, |
| "validation/probe_transcript_similarity_by_turn/turn_6": 0.20982533421592103, |
| "validation/goal_loss": 6.045462267592054 |
| }, |
| "global_step": 386, |
| "train_dataset": { |
| "example_count": 7696, |
| "pair_count_mean": 4.2548076923076925, |
| "pair_count_max": 6, |
| "response_target_tokens_mean_turn2_plus": 126.23921517671518, |
| "response_target_tokens_max_turn2_plus": 1643, |
| "probe_target_tokens_mean_turn2_plus": 471.31613825363826, |
| "probe_target_tokens_max_turn2_plus": 3285, |
| "dataset_counts": { |
| "chatalpaca_multiturn_enriched": 7696 |
| }, |
| "probe_question_text": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [Assistant 1]: Y [User 2]: A etc" |
| }, |
| "validation_dataset": { |
| "example_count": 151, |
| "pair_count_mean": 4.139072847682119, |
| "pair_count_max": 6, |
| "response_target_tokens_mean_turn2_plus": 123.21854304635761, |
| "response_target_tokens_max_turn2_plus": 889, |
| "probe_target_tokens_mean_turn2_plus": 448.5298013245033, |
| "probe_target_tokens_max_turn2_plus": 1545, |
| "dataset_counts": { |
| "chatalpaca_multiturn_enriched": 151 |
| }, |
| "probe_question_text": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [Assistant 1]: Y [User 2]: A etc" |
| }, |
| "wandb_enabled": true |
| } |
| ] |
| } |
