Logical Reasoning on RuleTaker (held-out)
62.63Performance (%)ENVISIONS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ENVISIONSBackbone=LLaMA2-Chat (7B), Approach=Env-guided Self-Training2024.06 | 62.63 | 58.69 | |
| ENVISIONSBackbone=LLaMA2-Chat (13B), Approach=Env-guided Self-Training2024.06 | 60.84 | 63.56 | |
| STaR + Env.Backbone=LLaMA2-Chat (13B), Approach=Env-guided Self-Training2024.06 | 60.33 | 60.6 | |
| Iterative SFT+DPOBackbone=LLaMA2-Chat (13B), Approach=Reinforced Self-Training2024.06 | 59.47 | 59.77 | |
| Iterative SFT+DPOBackbone=LLaMA2-Chat (7B), Approach=Reinforced Self-Training2024.06 | 58.82 | 54.73 | |
| STaR + Env.Backbone=LLaMA2-Chat (7B), Approach=Env-guided Self-Training2024.06 | 58.6 | 55.91 | |
| Self-RewardingBackbone=LLaMA2-Chat (13B), Approach=Reinforced Self-Training2024.06 | 56.66 | 52.5 | |
| LLaMA2-Chat (few-shot)Backbone=LLaMA2-Chat (13B), Mode=few-shot2024.06 | 51.5 | 38.61 | |
| GPT-4 + LLaMA2-ChatBackbone=LLaMA2-Chat (13B), Approach=Distill-then-Finetune2024.06 | 50.61 | 56.49 | |
| Self-RewardingBackbone=LLaMA2-Chat (7B), Approach=Reinforced Self-Training2024.06 | 50.04 | 44.22 | |
| Claude-2 + LLaMA2-ChatBackbone=LLaMA2-Chat (7B), Approach=Distill-then-Finetune2024.06 | 49.17 | 53.03 | |
| GPT-4 + LLaMA2-ChatBackbone=LLaMA2-Chat (7B), Approach=Distill-then-Finetune2024.06 | 48.88 | 52.46 | |
| Claude-2 + LLaMA2-ChatBackbone=LLaMA2-Chat (13B), Approach=Distill-then-Finetune2024.06 | 48.45 | 57.17 | |
| LLaMA2-Chat (few-shot)Backbone=LLaMA2-Chat (7B), Mode=few-shot2024.06 | 47.44 | 25.69 |