AI platform
-
Component of distributed model
- Master: manages other nodes
- Workers: works on the portion of the training
- Parameters server: coordinates shaded model state between worker
-
Types of predictions
- Online
- Real-time processing with fully managed ML Engine
- No Docker container required & supports multiple framework
- Batch
- For asynchronous operations
- Scales to terabytes of data
- Online
-
Terminolgy
- Model
- Version: instance of model
- Job: interaction with AI Platform
IAM Roles
- Project and Model roles
- Admin
- Developer
- Viewer
- Model Only roles
- Model Owner: full access to model and versions
- Model User:
- Read models and use for prediction
- Easy to share specific models
Machine scale
-
BASIC
-
STANDARD_1
-
PREMIUM_1
-
BASIC_GPU
-
CUSTOM
-
GPU(Graphic Processing Unit) /TPU (Tensor Processing Unit)