Likelihood-free Framework
-
New Model Design Aims to Cut High Enterprise AI Costs
A new architectural design, Continuous Autoregressive Language Models (CALM), offers potential cost savings for enterprises deploying AI. CALM predicts continuous vectors instead of discrete tokens, compressing information and reducing computational steps. Experiments show CALM models achieve comparable performance to baselines with significantly fewer FLOPs. This novel approach requires a new “likelihood-free framework” including training methods, a BrierLM evaluation metric, and a likelihood-free sampling algorithm. CALM highlights a shift towards architectural efficiency as a crucial factor in reducing enterprise AI costs and improving sustainability.
