6 papers
RL's Razor: Why Online Reinforcement Learning Forgets Less
Idan Shenfeld, Jyothish Pari, Pulkit Agrawal
Comparison of fine-tuning models with reinforcement learning (RL) and supervised fine-tuning (SFT) reveals that, despite similar performance at a new task, RL preserves prior knowl…
Best-of-N through the Smoothing Lens: KL Divergence and Regret Analysis
Gholamali Aminian, Idan Shenfeld, Amir R. Asadi +2
A simple yet effective method for inference-time alignment of generative models is Best-of- (BoN), where outcomes are sampled from a reference policy, evaluated using a prox…
Language Model Personalization via Reward Factorization
Idan Shenfeld, Felix Faltings, Pulkit Agrawal +1
Modern large language models (LLMs) are optimized for human-aligned responses using Reinforcement Learning from Human Feedback (RLHF). However, existing RLHF approaches assume a un…
KL-Regularized RLHF with Multiple Reference Models: Exact Solutions and Sample Complexity
Gholamali Aminian, Amir R. Asadi, Idan Shenfeld +1
Recent methods for aligning large language models (LLMs) with human feedback predominantly rely on a single reference model, which limits diversity, model overfitting, and underuti…
Learning How Hard to Think: Input-Adaptive Allocation of LM Computation
Mehul Damani, Idan Shenfeld, Andi Peng +2
Computationally intensive decoding procedures--including search, reranking, and self-critique--can improve the quality of language model (LM) outputs in problems spanning code gene…
The Future of Open Human Feedback
Shachar Don-Yehiya, Ben Burtenshaw, Ramon Fernandez Astudillo +17
Human feedback on conversations with language language models (LLMs) is central to how these systems learn about the world, improve their capabilities, and are steered toward desir…