Question Answering on HotPotQA Modified
27Average CorrectnessMulti-Answer RLVR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Multi-Answer RLVRLoss type=Multi Loss, Prompting strategy=Multi Prompt, k (Set size)=32026.03 | 27 | 100 | 544 | 19 | |
| Multi-Answer RLCRLoss type=Multi Loss, Prompting strategy=Multi Prompt, k (Set size)=32026.03 | 27 | 100 | 622 | 19 | |
| RLVRLoss type=Single Loss, Prompting strategy=Multi Prompt, k (Set size)=32026.03 | 23 | 100 | 551 | 17 | |
| RLCRLoss type=Single Loss, Prompting strategy=Multi Prompt, k (Set size)=32026.03 | 23 | 98 | 670 | 17 | |
| RLCRLoss type=Single Loss, Prompting strategy=Single Prompt, k (Set size)=32026.03 | 22 | 62 | 1,782 | 17 | |
| Zero-ShotPrompting strategy=Multi Prompt, Baseline group=RLVR, k (Set size)=32026.03 | 20 | 75 | 1,265 | 17 | |
| Zero-ShotPrompting strategy=Multi Prompt, Baseline group=RLCR, k (Set size)=32026.03 | 20 | 73 | 1,298 | 15 | |
| RLVRLoss type=Single Loss, Prompting strategy=Single Prompt, k (Set size)=32026.03 | 17 | 59 | 1,466 | 19 |