Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
SlideShare a Scribd company logo

Dockerize Spago Self Contained ML & NLP Library & Deploy on Okteto Cloud Using Okteto Stack

•
•

This document discusses using Docker for machine learning and natural language processing projects. It introduces Spago, an open source machine learning library written in Go, and demos deploying a Spago project to Kubernetes using Okteto. The key points are: - Docker is useful for ML development as it allows packaging models and code into containers for scalable and reproducible deployments. - Spago is a pure Go library for NLP tasks like named entity recognition and question answering that aims to optimize CPU usage. - Okteto allows easily deploying Docker Compose stacks to Kubernetes with minimal configuration. - A live demo then shows deploying a Spago project to Kubernetes with Okteto.

1 of 19
Download to read offline

1

Spago Self ConTained ML & NLP Library & Deploy on Okteto

2

Agenda For all level Audience ( Docker Knowledge + point ) Docker for Golang Developer ML + NLP Why Docker + ML Live Demo ! 2

3

“ 3 Sangam Biradar Docker Community Leader Cloud Native Developer . Github: Sangam14 Join Community http://discord.kubedaily.live/

4

Dockerize Spago Self Contained ML & NLP Library & Deploy on Okteto Cloud Using Okteto Stack

5

Why Docker + ML? Docker is Not just in ML but across the entire development phase

6

What you should & shouldn’t include in ML Development Container ⩥ Only the ML frameworks & Dependencies ⩥ ML frameworks and Dependencies & training code ( scaling workload on cluster ) 6

7

7 Spago A Machine Learning Library Written in Pure Go Designed to support Neural Architectures in NLP

8

- educational purpose - production use - High Quality Implementation Purpose of spago 8

9

Where would I use SpaGo ? 9 ⩥ Production NLP system ⩥ Required Optimization of CPU interference ⩥ Relies on GONUM assembly code with SIMD instruction ⩥ CGO is bottleneck , so there are experiment with intel MKL, Apache Arrow ⩥ Also experiment with sparse model , LSH , etc ⩥ Training of No-very-deep model ⩥ The training seems less interesting as its expensive for modern model even with GPUS/TPUs

10

Ready to use ⩥ Compatible with pytorch model ⩥ Transformer 50+ pretrained model in 100+ langauges 10

11

Use for “production level ml/nlp” ⩥ Train shallow models ⩥ fine tune deep model ⩥ inference 11 ⩥ Train deep model ⩥ fine tune deep model ⩥ inference Weight importer

12

Real time examples ⩥ Named Entities Recognition ⩥ Hugging Face Importer ⩥ Question Answering ⩥ Masked Language Model 12

13

Okteto – Kubernetes Developer platform 13

14

14

15

Docker Compose on kubernetes ⩥ okteto stack is easy way to deploy your docker compose within secs . 15

16

16 Image name (eg. any image from docker hub name of your stack build context Override default entry point Port exposed by each service Resource quota volume

17

Demo Time !! 17

18

Connect with me ! ⩥ Join Discord.kubedaily.live ⩥ Subscribe ! Kubedaily.live ⩥ #Learn Golang – Gopherlabs.kubedaily.com ⩥ #Container Technology : Containerlabs.kubedaily.com 18

19

19 Thanks ! subscribe !

More Related Content

Dockerize Spago Self Contained ML & NLP Library & Deploy on Okteto Cloud Using Okteto Stack

  • 1. Spago Self ConTained ML & NLP Library & Deploy on Okteto
  • 2. Agenda For all level Audience ( Docker Knowledge + point ) Docker for Golang Developer ML + NLP Why Docker + ML Live Demo ! 2
  • 3. “ 3 Sangam Biradar Docker Community Leader Cloud Native Developer . Github: Sangam14 Join Community http://discord.kubedaily.live/
  • 5. Why Docker + ML? Docker is Not just in ML but across the entire development phase
  • 6. What you should & shouldn’t include in ML Development Container ⩥ Only the ML frameworks & Dependencies ⩥ ML frameworks and Dependencies & training code ( scaling workload on cluster ) 6
  • 7. 7 Spago A Machine Learning Library Written in Pure Go Designed to support Neural Architectures in NLP
  • 8. - educational purpose - production use - High Quality Implementation Purpose of spago 8
  • 9. Where would I use SpaGo ? 9 ⩥ Production NLP system ⩥ Required Optimization of CPU interference ⩥ Relies on GONUM assembly code with SIMD instruction ⩥ CGO is bottleneck , so there are experiment with intel MKL, Apache Arrow ⩥ Also experiment with sparse model , LSH , etc ⩥ Training of No-very-deep model ⩥ The training seems less interesting as its expensive for modern model even with GPUS/TPUs
  • 10. Ready to use ⩥ Compatible with pytorch model ⩥ Transformer 50+ pretrained model in 100+ langauges 10
  • 11. Use for “production level ml/nlp” ⩥ Train shallow models ⩥ fine tune deep model ⩥ inference 11 ⩥ Train deep model ⩥ fine tune deep model ⩥ inference Weight importer
  • 12. Real time examples ⩥ Named Entities Recognition ⩥ Hugging Face Importer ⩥ Question Answering ⩥ Masked Language Model 12
  • 13. Okteto – Kubernetes Developer platform 13
  • 14. 14
  • 15. Docker Compose on kubernetes ⩥ okteto stack is easy way to deploy your docker compose within secs . 15
  • 16. 16 Image name (eg. any image from docker hub name of your stack build context Override default entry point Port exposed by each service Resource quota volume
  • 18. Connect with me ! ⩥ Join Discord.kubedaily.live ⩥ Subscribe ! Kubedaily.live ⩥ #Learn Golang – Gopherlabs.kubedaily.com ⩥ #Container Technology : Containerlabs.kubedaily.com 18