Search papers, labs, and topics across Lattice.
This paper introduces a reinforcement learning (RL) approach to train language models (LMs) to generate multiple plausible answers in a single forward pass, rather than collapsing the distribution to a single mode. The method modifies the RL objective to explicitly encourage the generation of diverse candidate answers, effectively internalizing inference-time search. Experiments across question-answering, medical diagnosis, and coding benchmarks demonstrate improved diversity, coverage, and calibration compared to single-answer baselines, with increased accuracy on coding tasks.
LMs can learn to generate multiple plausible answers in a single forward pass, outperforming traditional single-answer models on tasks requiring distributional reasoning and offering a compute-efficient alternative to best-of-k sampling.
Given a question, a language model (LM) implicitly encodes a distribution over possible answers. In practice, post-training procedures for LMs often collapse this distribution onto a single dominant mode. While this is generally not a problem for benchmark-style evaluations that assume one correct answer, many real-world tasks inherently involve multiple valid answers or irreducible uncertainty. Examples include medical diagnosis, ambiguous question answering, and settings with incomplete information. In these cases, we would like LMs to generate multiple plausible hypotheses, ideally with confidence estimates for each one, and without computationally intensive repeated sampling to generate non-modal answers. This paper describes a multi-answer reinforcement learning approach for training LMs to perform distributional reasoning over multiple answers during inference. We modify the RL objective to enable models to explicitly generate multiple candidate answers in a single forward pass, internalizing aspects of inference-time search into the model's generative process. Across question-answering, medical diagnostic, and coding benchmarks, we observe improved diversity, coverage, and set-level calibration scores compared to single answer trained baselines. Models trained with our approach require fewer tokens to generate multiple answers than competing approaches. On coding tasks, they are also substantially more accurate. These results position multi-answer RL as a principled and compute-efficient alternative to inference-time scaling procedures such as best-of-k. Code and more information can be found at https://multi-answer-rl.github.io/.
MIT CSAIL