Medical Long-form Generation on MedQA (test)
85SMAPCD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| APCDBackbone=II-Medical-8B-1706, Decoding Strategy=APCD2026.05 | 85 | 86.7 | |
| CSBackbone=II-Medical-8B-1706, Decoding Strategy=CS2026.05 | 83.4 | 85.8 | |
| Sample (top k + top p)Backbone=II-Medical-8B-1706, Decoding Strategy=Sample (top k + top p)2026.05 | 80.1 | 85.2 | |
| Beam SearchBackbone=II-Medical-8B-1706, Decoding Strategy=Beam Search2026.05 | 79.9 | 83.1 | |
| II-Medical-8B-1706Backbone=II-Medical-8B-1706, Decoding Strategy=Standard2026.05 | 79.2 | 82.8 | |
| DoLa (low)Backbone=II-Medical-8B-1706, Decoding Strategy=DoLa (low)2026.05 | 76.7 | 84.5 | |
| DBSBackbone=II-Medical-8B-1706, Decoding Strategy=DBS2026.05 | 72.3 | 82 | |
| DoLa (high)Backbone=II-Medical-8B-1706, Decoding Strategy=DoLa (high)2026.05 | 12.3 | 12.7 |