1: \begin{abstract}

2: While recent preference alignment algorithms for language models have demonstrated promising results, supervised fine-tuning (SFT) remains imperative for achieving successful convergence. In this paper, we study the crucial role of SFT within the context of preference alignment, emphasizing that a minor penalty for the disfavored generation style is sufficient for preference-aligned SFT. Building on this foundation, we introduce a straightforward and innovative reference model-free monolithic odds ratio preference optimization algorithm, \texttt{ORPO}, eliminating the necessity for an additional preference alignment phase. We demonstrate, both empirically and theoretically, that the odds ratio is a sensible choice for contrasting favored and disfavored styles during SFT across the diverse sizes from 125M to 7B. Specifically, fine-tuning Phi-2 (2.7B), Llama-2 (7B), and Mistral (7B) with \texttt{ORPO} on the UltraFeedback alone surpasses the performance of state-of-the-art language models with more than 7B and 13B parameters: achieving up to 12.20\% on $\text{AlpacaEval}_{2.0}$ (Figure \ref{fig:first}), 66.19\% on IFEval (instruction-level loose, Table \ref{tab:ifeval}), and 7.32 in MT-Bench (Figure \ref{fig:mtbench}). We release code\footnote{\url{https://github.com/xfactlab/orpo}} and model checkpoints for Mistral-\texttt{ORPO}-$\alpha$ (7B)\footnote{\url{https://huggingface.co/kaist-ai/mistral-orpo-alpha}} and Mistral-\texttt{ORPO}-$\beta$ (7B).\footnote{\url{https://huggingface.co/kaist-ai/mistral-orpo-beta}}

3: \end{abstract}

4: