Bayesian Code Diffusion for Efficient Automatic Deep Learning Program Optimization

We introduce Bayesian code diffusion, a new deep learning program optimization strategy devised to accelerate the autotuning process of deep learning compilers. Using the concepts of prior and posterior distributions in the Bayesian framework and reformulating them in the context of deep learning program optimization, the proposed approach efficiently searches for optimal program code in a significantly reduced search space through an iterative diffusion of program code. To further enhance the efficiency of program optimization, we propose pre-training and fine-tuning for the cost model, which improves both the model's predictive accuracy and training efficiency. We implement Bayesian code diffusion in Ansor and evaluate its performance on a wide range of deep learning models on both CPUs and GPUs. Existing approaches struggle to reliably generate high-performing deep learning programs, i.e. achieving low program execution latency, across various configurations, including diverse deep learning model architectures and hardware platforms (CPU and GPU). In contrast, Bayesian code diffusion reduces the end-to-end compilation (optimization) time required to generate the equivalent program execution latency in various configurations, i.e., achieving up to 3.31× optimization speedup. This substantial improvement demonstrates that Bayesian code diffusion performs efficient and principled deep learning program optimization across a wide range of deep learning models, operators, and hardware (CPU and GPU).