Soumith Chintala Facebook AI Research Overview What is Torch? The - - PowerPoint PPT Presentation

▶
soumith chintala facebook ai research overview
SMART_READER_LITE
LIVE PREVIEW

Soumith Chintala Facebook AI Research Overview What is Torch? The - - PowerPoint PPT Presentation

Aug 19, 2023 162 likes •473 views

Growing a research platform for cutting edge AI Soumith Chintala Facebook AI Research Overview What is Torch? The Community Common use Core Philosophy Key drivers of adoption Questions What is ? Interactive

slide-1
SLIDE 1

Growing a research platform for cutting edge AI

Soumith Chintala Facebook AI Research

slide-2
SLIDE 2

Overview

  • What is Torch?
  • The Community
  • Common use
  • Core Philosophy
  • Key drivers of adoption
  • Questions
slide-3
SLIDE 3

What is ?

  • Interactive Scientific computing framework
slide-4
SLIDE 4

What is ?

  • Interactive Scientific computing framework
slide-5
SLIDE 5

What is ?

  • Similar to Matlab / Python+Numpy
slide-6
SLIDE 6

What is ?

  • Little language overhead compared to Python / Matlab
  • JIT compilation via LuaJIT
  • Fearlessly write for-loops

Code snippet from a core package

slide-7
SLIDE 7

What is ?

  • Easy integration into and from C
  • Example: using CuDNN functions
slide-8
SLIDE 8

What is ?

  • Strong GPU support
slide-9
SLIDE 9

Community

slide-10
SLIDE 10

Community

slide-11
SLIDE 11

Community

slide-12
SLIDE 12

Community

slide-13
SLIDE 13

Community

slide-14
SLIDE 14

Community

slide-15
SLIDE 15

Community

slide-16
SLIDE 16

Community

slide-17
SLIDE 17

Community

slide-18
SLIDE 18

Community

slide-19
SLIDE 19

Community

slide-20
SLIDE 20

Neural Networks

  • nn: neural networks made easy
  • building blocks of differentiable modules
slide-21
SLIDE 21

Advanced Neural Networks

  • nngraph
  • easy construction of complicated neural networks
slide-22
SLIDE 22

autograd by

  • Write imperative programs
  • Backprop defined for every operation in the language
slide-23
SLIDE 23

Distributed Learning

  • in-built multi-GPU (data and model parallel)
  • distlearn by
  • multi-node parallelism
slide-24
SLIDE 24

Core Philosophy

  • Interactive computing
  • No compilation time
  • Imperative programming
  • Write code like you always did, not computation graphs in a

hacked up DSL

  • Minimal abstraction
  • Thinking linearly
  • Maximal Flexibility
  • No constraints on interfaces or classes
slide-25
SLIDE 25

There is no silver bullet

Slide credit: Yangqing Jia

Industry: Stability Scale & speed Data Integration Relatively Fixed Research:
 Flexible Fast Iteration Debuggable Relatively bare bone

Caffe Torch Theano TensorFlow D4J etc. Neon

slide-26
SLIDE 26

There is no silver bullet

Slide credit: Yangqing Jia

Industry: Stability Scale & speed Data Integration Relatively Fixed Research:
 Flexible Fast Iteration Debuggable Relatively bare bone

Caffe Torch Theano TensorFlow D4J etc. Neon

slide-27
SLIDE 27

There is no silver bullet

Slide credit: Yangqing Jia

Industry: Stability Scale & speed Data Integration Relatively Fixed Research:
 Flexible Fast Iteration Debuggable Relatively bare bone

Caffe Torch Theano TensorFlow D4J etc. Neon

slide-28
SLIDE 28

There is no silver bullet

Slide credit: Yangqing Jia

Industry: Stability Scale & speed Data Integration Relatively Fixed Research:
 Flexible Fast Iteration Debuggable Relatively bare bone

Caffe Torch Theano TensorFlow D4J etc. Neon

slide-29
SLIDE 29

Key Drivers of Adoption

  • Tutorials and support
  • Pre-trained models
  • High-quality open-source projects
  • Deeply integrated GPU goodness
  • Minimal abstractions
  • Imperative programming
  • Zero compile-time
slide-30
SLIDE 30

Questions!