| { |
| "phase_results": [ |
| { |
| "phase": "simple_multiturn_transcript", |
| "best_selection_metric_value": 2.7354467349895404, |
| "best_metrics": { |
| "validation/loss_total": 1.7354467349895404, |
| "validation/loss_response": 2.3269009419351203, |
| "validation/loss_current_user_reconstruction": 0.0, |
| "validation/loss_probe": 0.0, |
| "validation/loss_past_response_contrastive": 0.35538691878318784, |
| "validation/response_first_token_exact_match": 0.3645833333333333, |
| "validation/current-user_reconstruction_first_token_exact_match": 0.0, |
| "validation/probe_first_token_exact_match": 0.0, |
| "validation/supervised_turn_count": 96.0, |
| "validation/response_exact_match": 0.0, |
| "validation/response_similarity": 0.10222400877569102, |
| "validation/response_token_f1": 0.20444801755138203, |
| "validation/response_line_recall": 0.0, |
| "validation/response_reconstruction_similarity": 0.0, |
| "validation/response_reconstruction_exact_match": 0.0, |
| "validation/response_reconstruction_token_f1": 0.0, |
| "validation/probe_exact_match": 0.0, |
| "validation/probe_transcript_similarity": 0.0, |
| "validation/probe_token_f1": 0.0, |
| "validation/probe_line_recall": 0.0, |
| "validation/response_similarity_by_turn/turn_2": 0.13293650793650794, |
| "validation/response_similarity_by_turn/turn_3": 0.1046455938697318, |
| "validation/response_similarity_by_turn/turn_4": 0.10050017163422015, |
| "validation/response_similarity_by_turn/turn_5": 0.07827223148007878, |
| "validation/response_similarity_by_turn/turn_6": 0.09476553895791645, |
| "validation/goal_loss": 2.7354467349895404 |
| }, |
| "global_step": 990, |
| "train_dataset": { |
| "example_count": 7904, |
| "pair_count_mean": 6.044534412955466, |
| "pair_count_max": 7, |
| "response_target_tokens_mean_turn2_plus": 232.00544028340082, |
| "response_target_tokens_max_turn2_plus": 1171, |
| "probe_target_tokens_mean_turn2_plus": 1124.4765941295548, |
| "probe_target_tokens_max_turn2_plus": 4800, |
| "dataset_counts": { |
| "chatalpaca_multiturn_enriched": 7904 |
| }, |
| "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", |
| "current_user_reconstruction_question_text": "What did me (the user) just ask you? give verbatim message I just previously sent" |
| }, |
| "validation_dataset": { |
| "example_count": 40, |
| "pair_count_mean": 6.125, |
| "pair_count_max": 7, |
| "response_target_tokens_mean_turn2_plus": 250.925, |
| "response_target_tokens_max_turn2_plus": 611, |
| "probe_target_tokens_mean_turn2_plus": 1229.375, |
| "probe_target_tokens_max_turn2_plus": 2447, |
| "dataset_counts": { |
| "chatalpaca_multiturn_enriched": 40 |
| }, |
| "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", |
| "current_user_reconstruction_question_text": "What did me (the user) just ask you? give verbatim message I just previously sent" |
| }, |
| "wandb_enabled": true |
| } |
| ] |
| } |
