Closed book Question Answering on MultiSpanQA (test)
48F1 ScoreCoVe (factored)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CoVe (factored)LLM=Llama 65B, Strategy=CoVe (factored)2023.09 | 48 | 50 | 46 | |
| CoVe (joint)LLM=Llama 65B, Strategy=CoVe (joint)2023.09 | 46 | 50 | 42 | |
| Llama 65B (Few-shot)LLM=Llama 65B, Strategy=Few-shot2023.09 | 39 | 40 | 38 | |
| Llama 2 70B Chat (Zero-shot)LLM=Llama 2 70B Chat, Strategy=Zero-shot2023.09 | 20 | 13 | 40 | |
| Llama 2 70B Chat (CoT)LLM=Llama 2 70B Chat, Strategy=CoT2023.09 | 17 | 11 | 37 |