Dockerize Spago Self Contained ML & NLP Library & Deploy on Okteto Cloud Using Okteto Stack
•
0 likes•140 views
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
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
More Related Content
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
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
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