List-based question answering on Wikidata Easier (test)
36PrecisionCoVe (two-step)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CoVe (two-step)LLM=Llama 65B2023.09 | 36 | 38 | 0.68 | |
| CoVe (factored)LLM=Llama 65B2023.09 | 32 | 38 | 0.79 | |
| CoVe (joint)LLM=Llama 65B2023.09 | 29 | 41 | 0.98 | |
| Few-shotLLM=Llama 65B2023.09 | 17 | 59 | 2.95 | |
| Zero-shotLLM=Llama 2 70B Chat2023.09 | 12 | 55 | 3.93 | |
| CoTLLM=Llama 2 70B Chat2023.09 | 8 | 75 | 8.92 |