Publications

TVM is developed as part of peer-reviewed research in machine learning compiler framework for CPUs, GPUs, and machine learning accelerators.

This document includes references to publications describing the research, results, and design that use or built on top of TVM.

2018

• TVM: An Automated End-to-End Optimizing Compiler for Deep Learning, [Slides]

• Learning to Optimize Tensor Programs, [Slides]

2020

• Ansor: Generating High-Performance Tensor Programs for Deep Learning, [Slides] [Tutorial]

2021

• Nimble: Efficiently Compiling Dynamic Neural Networks for Model Inference, [Slides]

• Cortex: A Compiler for Recursive Deep Learning Models, [Slides]

• UNIT: Unifying Tensorized Instruction Compilation, [Slides]

• Lorien: Efficient Deep Learning Workloads Delivery, [Slides]

• Bring Your Own Codegen to Deep Learning Compiler, [Slides] [Tutorial]

2022

• DietCode: Automatic optimization for dynamic tensor program, [Slides]

• Bolt: Bridging the Gap between Auto-tuners and Hardware-native Performance, [Slides]

• The CoRa Tensor Compiler: Compilation for Ragged Tensors with Minimal Padding, [Slides]