Apache MXNet 2.0: Towards Synergistic ML and DL with Standardization

on YouTube ↗30m

" summary

Apache MXNet 2.0 standardizes on the Python Array API and Open Neural Network Exchange (ONNX), adds numpy-compatible np/mpx arrays with automatic differentiation and GPU support, and surfaces Gluon 2.0 for hybridized, serializable models. It implements a DLPack zero-copy interchange for async device transfers, supports custom C++ operators, TVM and custom backends, and cites MLPerf 1.0 results where MXNet appears in 30 of 67 submissions and Intel oneDNN speedups (Mobilenet ~500x, ResNet50 ~800x). Community counts: ~960 contributors, ~20k stars, ~7k forks.

AI-written summary · about

" chapters (12)

1. 0:00Start
2. 3:07Different Design Choices
3. 8:45Fragmentation is costly
4. 10:25The Remedy
5. 10:44Open Neural Network Exchange
6. 11:12Consortium for Python Data API Standards
7. 12:15Array API Standard
8. 12:52Out of Scope and Non-goals • Out of scope - Implementation of the standard - Execution semantics
9. 14:17Array API Data Interchange
10. 16:39Apache MXNet 2.0
11. 23:25Community
12. 28:14Sockeye 2.3

" related talks

• 29mAccelerate NumPy Data Science Workloads and Deep Learning ApplicationsHao Jin · PyData Los Angeles 2019
• 39mAccelerate large-scale machine learning with NP on MXNetHao Jin · PyData Austin 2019
• 42mAnima Anadkumar, Principal Scientist, Amazon Web Services, Endowed Professor, CalTechAnima Anadkumar · MLconf SF 2017
• 20mApache MXNet, a highly memory efficient deep learning frameworkGirish Patil · Anthill Inside 2017
• 30mAnyone can Build Great Deep Learning Applications - Deep NumpyCyrus Vahid · PyData Warsaw 2019
• 41mOpen standards for machine learning deploymentNick Pentreath · Berlin Buzzwords 2019
• 38mThe new kid on the block: deep learning with GluonNLPSneha Jha · PyData NYC 2018
• 01h 16mDeep Dive into Deep Learning with MXNetRandall Hunt · Devoxx Belgium 2017