Multi-hop retrieval question answering on HotPotQA (test)
45.6Answer Accuracymultihop
Evaluation results
| Method | Links | ||
|---|---|---|---|
| multihopModel=GPT-3.5, Compiler=ensemble, eval_subset=50% of test set2023.10 | 45.6 | — | |
| multihopModel=Llama2-13b-chat, Compiler=ensemble2023.10 | 41 | — | |
| multihopModel=GPT-3.5, Compiler=bootstrap2023.10 | 39.6 | 43.8 | |
| multihopModel=Llama2-13b-chat, Compiler=bootstrap2023.10 | 36.4 | 43.5 | |
| COT_RAGModel=Llama2-13b-chat, Compiler=bootstrap2023.10 | 32.9 | 34.4 | |
| vanillaModel=GPT-3.5, Compiler=fewshot2023.10 | 31.5 | — | |
| multihopModel=Llama2-13b-chat, Compiler=fewshot2023.10 | 31.3 | 30.8 | |
| multihopModel=GPT-3.5, Compiler=fewshot2023.10 | 31.2 | 40.8 | |
| COT_RAGModel=GPT-3.5, Compiler=fewshot2023.10 | 29.8 | 34.4 | |
| COT_RAGModel=Llama2-13b-chat, Compiler=fewshot2023.10 | 28 | 34.4 | |
| vanillaModel=Llama2-13b-chat, Compiler=fewshot2023.10 | 21.8 | — |