Open-ended Question Answering on NarrativeQA (Acc, BLEU, F1)
73.77Accuracyw/t BoT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| w/t BoTModel=Qwen2.5-1.5b-instruct, Geometric Component=Barycentric Transport2026.05 | 73.77 | 54.85 | 73.82 | |
| w/t CDModel=Qwen2.5-1.5b-instruct, Geometric Component=Cosine Dispersion2026.05 | 72.59 | 53.72 | 73.25 | |
| w/t KLEModel=Qwen2.5-1.5b-instruct2026.05 | 72.03 | 52.11 | 72.39 | |
| w/t SEModel=Qwen2.5-1.5b-instruct2026.05 | 71.76 | 52.2 | 72.24 | |
| CAREModel=Qwen2.5-1.5b-instruct2026.05 | 71.4 | 51.84 | 71.65 | |
| GRPOModel=Qwen2.5-1.5b-instruct2026.05 | 66.23 | 48.71 | 67.02 | |
| EMPOModel=Qwen2.5-1.5b-instruct2026.05 | 57.38 | 41.32 | 60.96 |