GLOW - AshokBhat/ml GitHub Wiki

• ML Compiler for hardware accelerators
• FB uses ML specific ASICs, which require a specialized compiler to translate the graph to instructions that execute on these accelerators.
• https://github.com/pytorch/glow
• No plans to use MLIR

• Input - Computational graph of PyTorch and other frameworks
• Output - Highly optimized code for machine learning accelerators.
• Optimizations - Domain and hardware-specific optimizations like kernel fusion

• Glow: Graph Lowering Compiler Techniques for Neural Networks
• https://arxiv.org/pdf/1805.00907.pdf

• PyTorch
• TVM | XLA | GLOW
• MLIR