Question Answering on FinQA (val)
0.6122Execution AccuracyFinQANet
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FinQANetBackbone=RoBerta-large2021.09 | 0.6122 | 0.5805 | |
| FinQANetBackbone=RoBerta-base2021.09 | 0.5627 | 0.5349 | |
| FinQANetBackbone=BERT-large2021.09 | 0.5386 | 0.5095 | |
| FinQANetBackbone=BERT-base2021.09 | 0.4991 | 0.4715 | |
| Retriever + NeRdBackbone=BERT-base2021.09 | 0.4753 | 0.4537 | |
| FinQANetBackbone=FinBert2021.09 | 0.4664 | 0.4411 | |
| Pre-Trained LongformerBackbone=base2021.09 | 0.2383 | 0.2256 | |
| Retriever + Seq2seq2021.09 | 0.1876 | 0.1752 | |
| TF-IDF + Single Op2021.09 | 0.0165 | 0.0165 | |
| Retriever + Direct Generation2021.09 | 0.0087 | — |