Cheaply Estimating Inference Efficiency Metrics for Autoregressive Transformer Models

Dec-26-2025, 20:39:40 GMT–Neural Information Processing Systems 

Large language models (LLMs) are highly capable but also computationally expensive.