LLM Alignment on AlpacaEval Length-Controlled (test)
8.78LC Win RateUNA-score (MSE)
Evaluation Results
| Method | Links | |
|---|---|---|
| UNA-score (MSE)Base Model=Mistral, Fine-tuning data=HelpSteer22024.08 | 8.78 | |
| UNA-score (MSE)Base Model=Llama, Fine-tuning data=HelpSteer22024.08 | 7.87 | |
| UNA-binary (BCE)Base Model=Mistral, Fine-tuning data=HelpSteer22024.08 | 7.41 | |
| KTOBase Model=Mistral, Fine-tuning data=HelpSteer22024.08 | 4.46 | |
| KTOBase Model=Llama, Fine-tuning data=HelpSteer22024.08 | 4.17 | |
| UNA-binary (BCE)Base Model=Llama, Fine-tuning data=HelpSteer22024.08 | 3.96 | |
| DPOBase Model=Mistral, Fine-tuning data=HelpSteer22024.08 | 3.67 | |
| UNA-pairwiseBase Model=Mistral, Fine-tuning data=HelpSteer22024.08 | 3.67 | |
| DPOBase Model=Llama, Fine-tuning data=HelpSteer22024.08 | 2.09 | |
| UNA-pairwiseBase Model=Llama, Fine-tuning data=HelpSteer22024.08 | 2.09 | |
| MistralBase Model=Mistral, Fine-tuning data=HelpSteer22024.08 | 0.31 | |
| LlamaBase Model=Llama, Fine-tuning data=HelpSteer22024.08 | 0.25 |