OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stopping

Large Language Models (LLMs) have achieved remarkable success in complex reasoning tasks through Chain-of-Thought (CoT) prompting. However, these models often exhibit "computational overthinking," generating redundant reasoning steps that increase latency and cost without improving accuracy. Recent studies suggest that CoT trajectories can be significantly pruned, yet existing methods often rely on forcing a static thinking budget, heuristic filtering, sub-optimal early exit via classification, or expensive re-training. In this paper, we introduce OS-Pruner, a lightweight plug-in framework that formulates chain-of-thought pruning as an optimal stopping problem. Given a reasoning prefix, OS-Pruner learns whether further reasoning is worth its token cost by optimizing an explicit utility that trades off final-answer accuracy against generated length. Our novel formulation enables the model to dynamically assess the sufficient point of termination for a reasoning chain. OS-Pruner is designed to be lightweight during both training and inference, and to provide users with fine-grained control over the reasoning-effort vs. accuracy trade-off. On diverse reasoning benchmarks and base models, OS-Pruner achieves 20-60\% reduction in generation length with minimal accuracy sacrifice.

CoDream: Exchanging dreams instead of models for federated aggregation with heterogeneous models

Federated Learning (FL) enables collaborative optimization of machine learning models across decentralized data by aggregating model parameters. Our approach extends this concept by aggregating "knowledge" derived from models, instead of model parameters. We present a novel framework called CoDream, where clients collaboratively optimize randomly initialized data using federated optimization in the input data space, similar to how randomly initialized model parameters are optimized in FL. Our key insight is that jointly optimizing this data can effectively capture the properties of the global data distribution. Sharing knowledge in data space offers numerous benefits: (1) model-agnostic collaborative learning, i.e., different clients can have different model architectures; (2) communication that is independent of the model size, eliminating scalability concerns with model parameters; (3) compatibility with secure aggregation, thus preserving the privacy benefits of federated learning; (4) allowing of adaptive optimization of knowledge shared for personalized learning. We empirically validate CoDream on standard FL tasks, demonstrating competitive performance despite not sharing model parameters. Our code: https://mitmedialab.github.io/codream.github.io/