This repo gives you hints and demos to jumpstart on vertex ai.
cloud_orbit_demo_1_apis.ipynb
explain the basic vertex ai apis
- vision.image
- sentiment_analysis
- entities recognition (nlp)
- google translate
cloud_orbit_demo_2_automl_endpoint_fetch.ipynb
explain the usage of automl.
- fetching answer from a deployed endpoint
cloud_orbit_demo_3_vertex_pipeline.ipynb
explain vertex ai pipelines
- creating components
- creating a pipeline with the components
- compiling the pipeline
- run a pipeline job
- use a yaml file to do programatic pipelines
- creating a pipeline for machine learning
- creating a tabular dataset based on big query
- creating an auto-ml model
- callling the previous component to know if the model's performance is good enough
- using a conditional test from "dsl" library to decide or not to deploy
- creating an endpoint to serve the model
- visualizing the pipeline in vertex's user interface :
cloud_orbit_demo_4_gpu_vs_cpu.ipynb
explore gpu calculus and endpoint deploiement
- testing cpu vs gpu calculus on vertex ai
- seeing that gpu is slower on small dataset, but faster in big ones
- creating a pipeline for data preparation
- create a pipeline model
- create a pipeline api endpoint with the created model
- testing the created endpoint
video_transcription_demo_github.ipynb
explain how to do an automatic summary of a video, with timestamped parts.
Example of result here
- import and upload a video in a bucket
- use vidéo transcription for speech to text
- use genai and prompt-engineering with text-bison@002 to do a chapitrage
- output as html