Intelligent Frontends - Angular and Transformers.js

Slide 1

Slide 1 text

I N T E L L I G E N T F R O N T E N D S A N G U L A R A N D T R A N S F O R M E R S . J S @markusingvarssn

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

No content

Slide 4

Slide 4 text

D E L H I A I R P O R T

Slide 5

Slide 5 text

S O M E R O U N D A B O U T @ D E L H I A I R P O R T

Slide 6

Slide 6 text

No content

Slide 7

Slide 7 text

I N T E L L I G E N T F R O N T E N D S A N G U L A R A N D T R A N S F O R M E R S J S After this talk, we will know about:

Slide 8

Slide 8 text

I N T E L L I G E N T F R O N T E N D S A N G U L A R A N D T R A N S F O R M E R S J S After this talk, we will know about: • Why Client-side Machine Learning?

Slide 9

Slide 9 text

I N T E L L I G E N T F R O N T E N D S A N G U L A R A N D T R A N S F O R M E R S J S After this talk, we will know about: • Why Client-side Machine Learning? • Transformers.js

Slide 10

Slide 10 text

I N T E L L I G E N T F R O N T E N D S A N G U L A R A N D T R A N S F O R M E R S J S After this talk, we will know about: • Why Client-side Machine Learning? • Transformers.js • Adding Transformers.js to our Angular projects

Slide 11

Slide 11 text

HELLO! INTELLIGENT F RONTENDS I’m Markus @markusingvarssn

Slide 12

Slide 12 text

HELLO! I’m Markus @markusingvarssn INTELLIGENT F RONTENDS

Slide 13

Slide 13 text

HELLO! I’m Markus @markusingvarssn INTELLIGENT F RONTENDS

Slide 14

Slide 14 text

M A C H I N E L E A R N I N G I N 3 0 S E C O N D S

Slide 15

Slide 15 text

M A C H I N E L E A R N I N G I N 3 0 S E C O N D S

Slide 16

Slide 16 text

M A C H I N E L E A R N I N G I N 3 0 S E C O N D S f(x) Dog Cat

Slide 17

Slide 17 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ?

Slide 18

Slide 18 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay!

Slide 19

Slide 19 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay! https://www.keycdn.com/support/network-latency

Slide 20

Slide 20 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay! https://www.nvidia.com/sv-se/geforce/graphics-cards/

Slide 21

Slide 21 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay! https://www.nvidia.com/sv-se/geforce/graphics-cards/ https://en.wikipedia.org/wiki/WebGPU

Slide 22

Slide 22 text

W E BGP U BR OWS E R S UP P ORT https://caniuse.com/webgpu

Slide 23

Slide 23 text

https://github.com/tensorflow/tfjs- models/tree/master/hand-pose-detection

Slide 24

Slide 24 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay! - It is private (and therefore supported offline!)

Slide 25

Slide 25 text

No content

Slide 26

Slide 26 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay! - It is private (and therefore supported offline!) - It is cost effective!

Slide 27

Slide 27 text

No content

Slide 28

Slide 28 text

W H Y C L I E N T- S I D E M A C H I N E L E A R N I N G ? - No network delay! - It is private (and therefore supported offline!) - It is cost effective!

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

No content

Slide 31

Slide 31 text

W H AT I S T R A N F O R M E R S . J S ? - Run pre-trained, state-of-the-art machine learning models in JavaScript

Slide 32

Slide 32 text

W H AT I S T R A N F O R M E R S . J S ? - Run pre-trained, state-of-the-art machine learning models in JavaScript - Mirrors the Python transformers library

Slide 33

Slide 33 text

W H AT I S T R A N F O R M E R S . J S ? - Run pre-trained, state-of-the-art machine learning models in JavaScript - Mirrors the Python transformers library - Supports 100+ model architectures across a wide domain of tasks

Slide 34

Slide 34 text

I N S TA L L I N G T R A N S F O R M E R S . J S

Slide 35

Slide 35 text

U S I N G T H E P I P E L I N E A P I

Slide 36

Slide 36 text

U S I N G T H E P I P E L I N E A P I A pipeline in Transformers.js is a high-level abstract that bundles together:

Slide 37

Slide 37 text

U S I N G T H E P I P E L I N E A P I A pipeline in Transformers.js is a high-level abstract that bundles together: - Pretrained model

Slide 38

Slide 38 text

U S I N G T H E P I P E L I N E A P I A pipeline in Transformers.js is a high-level abstract that bundles together: - Pretrained model - Input preprocessing

Slide 39

Slide 39 text

U S I N G T H E P I P E L I N E A P I A pipeline in Transformers.js is a high-level abstract that bundles together: - Pretrained model - Input preprocessing - Output postprocessing

Slide 40

Slide 40 text

A T I N Y E X A M P L E

Slide 41

Slide 41 text

A T I N Y E X A M P L E

Slide 42

Slide 42 text

T R A N S F O R M E R S . J S - TA S K S Natural Language Processing - summarization - question-answering - sentence-similarity - sentiment-analysis - text-generation - translation - etc Vision - background-removal - depth-estimation - image-classification - image-segmentation - object-detection - etc Audio - audio-classification - automatic-speech- recognition - text-to-speech - etc … and more

Slide 43

Slide 43 text

T R A N S F O R M E R S . J S - M O D E L S

Slide 44

Slide 44 text

T R A N S F O R M E R S . J S - M O D E L S

Slide 45

Slide 45 text

T R A N S F O R M E R S . J S - M O D E L S

Slide 46

Slide 46 text

M O R E E X A M P L E S ?

Slide 47

Slide 47 text

M O R E E X A M P L E S ?

Slide 48

Slide 48 text

M O R E E X A M P L E S ?

Slide 49

Slide 49 text

M O R E E X A M P L E S ?

Slide 50

Slide 50 text

M O R E E X A M P L E S ?

Slide 51

Slide 51 text

M O R E E X A M P L E S ?

Slide 52

Slide 52 text

M O R E E X A M P L E S ?

Slide 53

Slide 53 text

M O R E E X A M P L E S ?

Slide 54

Slide 54 text

L E T ’ S B U I L D S O M E T H I N G I N A N G U L A R !

Slide 55

Slide 55 text

S C R E E N D U M P F R O M M Y T H E S I S D A Y S

Slide 56

Slide 56 text

No content

Slide 57

Slide 57 text

L E T ’ S B U I L D A N O T H E R N O T E T A K I N G A P P

Slide 58

Slide 58 text

T O D AY ’ S A P P - O V E R V I E W • Record audio from user’s device • Use a speech-to-text model from Transformers.js to transcribe the recording to create rich notes

Slide 59

Slide 59 text

PROJECT STRUCTURE

Slide 60

Slide 60 text

APP.COMPONENT.TS

Slide 61

Slide 61 text

APP.COMPONENT.TS

Slide 62

Slide 62 text

APP.COMPONENT.TS

Slide 63

Slide 63 text

No content

Slide 64

Slide 64 text

APP.COMPONENT.TS

Slide 65

Slide 65 text

AUDIO-RECORED.COMPONENT.HTML

Slide 66

Slide 66 text

APP.COMPONENT.TS

Slide 67

Slide 67 text

APP.COMPONENT.TS

Slide 68

Slide 68 text

OTHER COMPONENTS

Slide 69

Slide 69 text

WORKER

Slide 70

Slide 70 text

What are Web Workers? • Runs JavaScript in a separate thread • Keeps UI smooth during heavy work

Slide 71

Slide 71 text

APP.WORKER.TS

Slide 72

Slide 72 text

WORKER.SERVICE.TS

Slide 73

Slide 73 text

WORKER.SERVICE.TS

Slide 74

Slide 74 text

WORKER.SERVICE.TS

Slide 75

Slide 75 text

WORKER.SERVICE.TS

Slide 76

Slide 76 text

WORKER.SERVICE.TS

Slide 77

Slide 77 text

INDEX.TS

Slide 78

Slide 78 text

TRANSCRIBER.SERVICE.TS

Slide 79

Slide 79 text

D E M O T I M E !

Slide 80

Slide 80 text

SUMMARY Thanks to Transformers.js and Angular, we can build ML-powered web apps that are

Slide 81

Slide 81 text

SUMMARY Thanks to Transformers.js and Angular, we can build ML-powered web apps that are Fast and responsive

Slide 82

Slide 82 text

SUMMARY Thanks to Transformers.js and Angular, we can build ML-powered web apps that are Fast and responsive Private and offline support

Slide 83

Slide 83 text

SUMMARY Thanks to Transformers.js and Angular, we can build ML-powered web apps that are Fast and responsive Private and offline-first Cost effective at scale

Slide 84

Slide 84 text

FUTURE WORK – WHAT’S NEXT?

Slide 85

Slide 85 text

FUTURE WORK – WHAT’S NEXT? • Train or fine-tune your own models!

Slide 86

Slide 86 text

FUTURE WORK – WHAT’S NEXT? • Train or fine-tune your own models! • Join the Huggingface community!

Slide 87

Slide 87 text

FUTURE WORK – WHAT’S NEXT? • Train or fine-tune your own models! • Join the Huggingface community! • Get creative!

Slide 88

Slide 88 text

A N G U L A R A N D T R A N S F O R M E R S . J S INTELLIGENT FRONTENDS Thank you! Markus Ingvarsson https://www.linkedin.com/in/markusingvarsson https://bsky.app/profile/markusingvarsson.bsky.social https://twitter.com/markusingvarssn http://if-ngindia.web.app/