https://bhavinjawade.github.io/tags/reward-modeling/ Recent content in Reward Modeling on Bhavin Jawade Hugo -- gohugo.io en-us Thu, 20 Feb 2025 00:00:00 +0000 https://bhavinjawade.github.io/post/orpo/ Thu, 20 Feb 2025 00:00:00 +0000 https://bhavinjawade.github.io/post/orpo/ Typically, preference alignment in large language models (LLMs) requires a reference model and a warm-up phase of supervised fine-tuning. ORPO proposes a monolithic approach by integrating preference alignment directly into the supervised fine-tuning (SFT) stage, ensuring that the model can learn human preferences during instruction tuning itself.