Facebook-Waston Liang-AI框架從解決研究到生產問題的探索： Facebook應對PyTorch部署挑戰的實踐.pdf

1、主辦方： 梁萬超 主辦方主辦方： Sergey Kareyev at Full Stack Deep Learning Bootcamp November 2019 主辦方主辦方： 主辦方主辦方： Ndarray library with acceleration Distributed transportJust-in-time compiler Utilities (data loading, etc) Gradient based optimization package Automatic differentiation engine Numpy alternative Deep le

2、arning, etc. 主辦方主辦方： PyTorch: An open source machine learning framework that accelerates the path from research prototyping to production deployment. Features with: Numpy-like ndarray (Tensor) Training with automatic differentiation Quick prototyping with Python Broad set of algorithms/apis to do de

3、ep learning Easy to extend Production ready 主辦方主辦方： 主辦方主辦方： 主辦方主辦方： We want a single tool for research and production Easier to transfer research results into models Diffi cult to do model conversion between frameworks Example like ONNX shows painful conversions between frameworks, especially between a dynamic and static framework Python proves much more easier to use. Flexibility is important. 主辦