LLM Alignment on HH-RLHF 300 prompts
69.8Win/Tie Rate vs Vanilla (GPT-4o)CARDS
Evaluation Results
| Method | Links | |
|---|---|---|
| CARDSModel=mistral-7b2025.11 | 69.8 | |
| CARDSModel=llama-7b2025.11 | 64.5 | |
| STARSModel=mistral-7b2025.11 | 64.5 | |
| DPOModel=mistral-7b2025.11 | 60.5 | |
| STARSModel=llama-7b2025.11 | 60.2 | |
| Tree-bonModel=mistral-7b2025.11 | 59.2 | |
| RAINModel=mistral-7b2025.11 | 59 | |
| ARGSModel=mistral-7b2025.11 | 58.8 | |
| DPOModel=llama-7b2025.11 | 56.4 | |
| Tree-bonModel=llama-7b2025.11 | 55.2 | |
| RAINModel=llama-7b2025.11 | 55 | |
| ARGSModel=llama-7b2025.11 | 54.8 | |
| Speculative-DecodingModel=mistral-7b2025.11 | 50.4 | |
| Speculative-DecodingModel=llama-7b2025.11 | 50.2 | |
| Vanilla LLMModel=llama-7b2025.11 | 50 | |
| Vanilla LLMModel=mistral-7b2025.11 | 50 |