Oct 16, 2025
EAGER: Entropy-Aware Generation for Adaptive Inference-Time Scaling
EAGer, a training-free method, reduces computation and improves performance by branching to multiple reasoning paths only when high-entropy tokens are present, reallocating compute budget to instances needing exploration, achieving up to 65% fewer tokens and 37% improvement in Pass@k.
Authors
Daniel Scalena, Leonidas Zotos, Elisabetta Fersini, Malvina Nissim, Ahmet Üstün
Abstract
With the rise of reasoning language models and test-time scaling methods as a paradigm for improving model performance, substantial computation is often required to generate multiple candidate sequences from the same prompt. This enables exploration of different reasoning paths toward the correct solution, however, allocates the same compute budget for each prompt. Grounded on the assumption that different prompts carry different degrees of complexity, and thus different computation needs, we propose EAGer, a training-free generation method that leverages model uncertainty through token-wise entropy distribution to reduce redundant computation and concurrently improve overall performance. EAGer allows branching to multiple reasoning paths only in the presence of high-entropy tokens, and then reallocates the saved compute budget to the instances where exploration of alternative paths is most needed. We find that across multiple open-source models on complex reasoning benchmarks such as AIME 2025, EAGer can reallocate the budget without accessing target labels, achieving the best efficiency-performance trade-off in terms of reasoning length and Pass@k. When target labels are accessible, EAGer generates up to 65% fewer tokens (hence saving compute) and achieves up to 37% improvement in Pass@k compared to the Full Parallel Sampling.
Related works
Research
Unlocking Reasoning Capability on Machine Translation in Large Language Models
Read
Research
Findings of the WMT25 Multilingual Instruction Shared Task: Persistent Hurdles in Reasoning, Generation, and Evaluation
Read
Research
CALIBER: Calibrating confidence before and after reasoning in language models
Read