New technique cuts LLM reasoning costs by 10x while keeping accuracy high

Original: Dual-Dimensional Consistency: Balancing Budget and Quality in Adaptive Inference-Time Scaling

Source: arxiv.org ↗