상세 컨텐츠

본문 제목

Advanced Testing Techniques

Generative AI with Large Language Models

by Taeyoon.Kim.DS 2023. 3. 7. 18:43

본문

 

Build Server

1. Run things

2. Open Source Server- Jenkins. Good product, configurable. Mobile. cons: Eventually server goes down. May not have the expertise to maintain the Jenkins nodes.

3rd party:

1. Circle CI

2. Github Actions - Have a Yaml file. Source control. Build server, ticket system, security scanning, ability to do a web hosting... etc There is more things you can get than just source controls. Using those features too. Code screening. including to the project.

Primary purpose: code quality then compelling offering. Don't need physical server. Efficient.

https://www.youtube.com/watch?v=7f9J8LV0lD8

CI/CD department. More options for the build servers.

Cloud - Native Build. What does it mean?

All cloud platforms offer. Competitng products with etc. (Data bricks? Server setup, )

AWS --> Code pipeline. Deeply Integrated. CD, there are some strong reason to use AWS Codepipe line.

GCP --> Cloud Build. Deep Integration. Automatically do a deployment. The key advantage of it.

AZURE -> DEVOps pipelines.

If on Cloud --> Use their native tools. CD tools. their Native tools. Azure same thing.

Potentially best strategy would be using both.

3rd party SAAS plays a role. powerful Dev tools. Reverage them, code analytics. Linting, Security test...

Github have the source control +

separate depolyment process.

For the analytics -> consistantly getting better. Actual CD aspect, Cloud native tools are good.

Source Control


Analytics - Github, Bitbucket, Gitlab

CloudNative - AWS Code commit, GCP, Azure... etc. Narrow path way. mirror your codes. native platforms. ups to the used case. DI, Speed. deploying your code frequently. Huge advantage. Pretty big one, security. Don't want to use security vector?

CI- Broken, Analytics tool.Statistics...

 

* python3 -m venv ~/.multicloud

* source ~/.multicloud/bin/activate

* which python

* git clone git@github.com:ewankim1023/multicloud.git --> Will be denied.

* ssh-keygen -t rsa

enter three times

* cat /home/ec2-user/.ssh/id_rsa.pub

ssh-rsa will be generated.

Go to Github settings -> adding ssh-rsa key in SSH. 

* git clone

 

Change Tabs: 4 not spaces for makefile

 

pylint --disable=R,C hello.py

disable some warnings. catch syntax errors easily.

 

Functional testing


Why? LoadTesting

SRC = Source Control

Linting, Testing are not enough to address the core problem wich is outages DB.

Staging envionment- duplication of what the production was.

Several different purformance test. Can I check in it an asset?

Dashboard --> Score board. 

 

Not doing functional testing, load testing.

1-2 years: Nothing works!!

 

Why CLI? Comand Line Interface


Why build software + How?

A) Solve problems. Web scraper, big data, ML etc

B) How? cut half complexity. Skeptical. 
By building a really small tool --> can to extremely complex can be done.

 

1. Ingest

2. Scrape

3. EDA

4. Prediction

 

Each piece does not care others, but can be tested. Then it's 10x more stronger. 

Build a small CLI first, average software developers do not do. How do I cut the complexity

How simple I can make it? Most projects fail because of complexity.

'Generative AI with Large Language Models' 카테고리의 다른 글

Generative AI & LLMs  (0) 2023.08.21
Introduction to LLMs and the generative AI project lifecycle  (0) 2023.08.21
Machine Learning pipeline  (0) 2023.03.13
AWS Innovate Data  (0) 2023.03.09
What is Testing?  (0) 2023.03.07

관련글 더보기