Presentation
A Holistic Functionalization Approach to Optimizing Imperative Tensor Programs in Deep Learning
DescriptionAs deep learning empowers various fields, many new operators have been proposed to improve the accuracy of deep learning models. Researchers often use imperative programming diagrams (PyTorch) to express these new operators, leaving the fusion optimization of these operators to deep learning compilers. Unfortunately, the inherent side effects introduced by imperative tensor programs, especially tensor-level mutations, often make optimization extremely difficult. We present a holistic functionalization approach (TensorSSA) to optimizing imperative tensor programs beyond control flow boundaries. We achieve a 1.79X (1.34X on average) speedup in representative deep learning tasks than state-of-the-art works.
Event Type
Research Manuscript
TimeTuesday, June 254:30pm - 4:45pm PDT
Location3001, 3rd Floor
AI
AI/ML Application and Infrastructure