covrl-variational-reasoning

Enhance language model reasoning through coupled sampling from prior (question-only) and posterior (answer-conditioned) distributions. Construct composite distribution mixing both at token level using hybrid sampling. Combine reconstruction term, selective NLL loss, and KL regularization. Achieve 12.4% improvement over base model and 2.3% over comparable ba…

Independent GitHub snapshot · Automated classification · Original documentation governs