Tag
The paper introduces Parallel Power Tempering (PPT), an inference-time sampling method that uses multiple parallel tempering chains at different sharpening levels to resolve the exploration-exploitation trade-off in power-sharpened LLM sampling, outperforming RL post-training and approaching frontier-model reasoning performance.
This paper presents BayesPO, a Bayesian prompt optimization framework using gradient-guided discrete MCMC with parallel tempering, achieving improved accuracy on instruction-induction tasks.