VitalDB - AnesLLM: a dataset and benchmark based on real intraoperative anaesthetic decisions 1.0
| scope | group | n_windows | top1_accuracy | top2_recall | result1_equals_aux | ratio_out_of_range | ratio_sum_gt_1 | invalid_real_vocab | invalid_result1_vocab | invalid_aux_vocab |
| split | overall | 84813 | 0.558 | 0.8043 | 0.0 | 0 | 0 | 0 | 0 | 0 |
| split | dev | 423 | 0.4941 | 0.7376 | 0.0 | 0 | 0 | 0 | 0 | 0 |
| split | train | 80933 | 0.5578 | 0.8043 | 0.0 | 0 | 0 | 0 | 0 | 0 |
| split | test | 3457 | 0.5684 | 0.8131 | 0.0 | 0 | 0 | 0 | 0 | 0 |
| difficulty | easy | 17359 | 0.615 | 0.8659 | 0.0 | 0 | 0 | 0 | 0 | 0 |
| difficulty | medium | 44852 | 0.5646 | 0.8094 | 0.0 | 0 | 0 | 0 | 0 | 0 |
| difficulty | hard | 22602 | 0.501 | 0.7469 | 0.0 | 0 | 0 | 0 | 0 | 0 |