| { |
| "phase_results": [ |
| { |
| "phase": "simple_multiturn_transcript", |
| "best_selection_metric_value": 6.619700041489555, |
| "best_metrics": { |
| "validation/loss_total": 2.042751923562315, |
| "validation/loss_response": 2.2427044281789876, |
| "validation/loss_current_user_reconstruction": 2.211277997021919, |
| "validation/loss_probe": 1.4743208408693655, |
| "validation/response_first_token_exact_match": 0.13291139240506328, |
| "validation/current-user_reconstruction_first_token_exact_match": 0.04008438818565401, |
| "validation/probe_first_token_exact_match": 0.6455696202531646, |
| "validation/supervised_turn_count": 474.0, |
| "validation/response_exact_match": 0.0, |
| "validation/response_similarity": 0.085281379048045, |
| "validation/response_token_f1": 0.17056275809609, |
| "validation/response_line_recall": 0.0, |
| "validation/response_reconstruction_similarity": 0.1664633586109287, |
| "validation/response_reconstruction_exact_match": 0.0, |
| "validation/response_reconstruction_token_f1": 0.1664633586109287, |
| "validation/probe_exact_match": 0.0, |
| "validation/probe_transcript_similarity": 0.1282942617309157, |
| "validation/probe_token_f1": 0.2565885234618314, |
| "validation/probe_line_recall": 0.0, |
| "validation/response_similarity_by_turn/turn_2": 0.08411492269405914, |
| "validation/response_similarity_by_turn/turn_3": 0.09334476082649767, |
| "validation/response_similarity_by_turn/turn_4": 0.073076306800475, |
| "validation/response_similarity_by_turn/turn_5": 0.08426057530291096, |
| "validation/response_similarity_by_turn/turn_6": 0.09150152785411145, |
| "validation/response_reconstruction_similarity_by_turn/turn_2": 0.21040282742381106, |
| "validation/response_reconstruction_similarity_by_turn/turn_3": 0.14500742221360224, |
| "validation/response_reconstruction_similarity_by_turn/turn_4": 0.1341406656584131, |
| "validation/response_reconstruction_similarity_by_turn/turn_5": 0.1833900922912523, |
| "validation/response_reconstruction_similarity_by_turn/turn_6": 0.1195707522775192, |
| "validation/probe_transcript_similarity_by_turn/turn_2": 0.1699551585606979, |
| "validation/probe_transcript_similarity_by_turn/turn_3": 0.14287379041502474, |
| "validation/probe_transcript_similarity_by_turn/turn_4": 0.09564061108210939, |
| "validation/probe_transcript_similarity_by_turn/turn_5": 0.06612852121879803, |
| "validation/probe_transcript_similarity_by_turn/turn_6": 0.0478720635643674, |
| "validation/goal_loss": 6.619700041489555 |
| }, |
| "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 |
| } |
| ] |
| } |
