Path1.Mod1.d - Explore ML Workspace - Regional Setups Flashcards
1
Q
Regional Training - Use Case
A
When you need to meet data compliance or have data movement constraints across regions.
2
Q
Lat Storage?
Regional Training - Pros
A
- Reduced network latency when training is done in the same data center as the data
- Supports Regional Storage attachment ; Regional Compute + attachable/detachable storage from different Regions (i.e. train with different data)
3
Q
Mgmt Res Quo
Regional Training - Cons
A
- Mgmt complexity when ML Pipelines run across multiple workspace instances
- Challenges in comparing experiment results across instances
- Overhead in quota and compute mgmt
4
Q
Regional Serving - Use Case
A
When you need to deploy closer to your target audience.
5
Q
Lat Inf Comp
Regional Serving Pros
A
- Minimized data latency and movement
- Inferencing in the data center where new data is ingested
- Compliance with local regulations
6
Q
Quo RFT
Regional Serving - Cons
A
- More overhead in quota and compute mgmt
- Difficulties with Regional fine tuning
7
Q
Regional Fine Tuning (Retraining) - Use Case
A
When you need to retrain a model (one initially trained with an initial data set) with Regional-specific data (Region-specific due to compliance or data movement constraints).
8
Q
Comp
Regional Fine Tuning (Retraining) - Pros
A
Experiment compliantly in data centers housing that data
9
Q
Res Quo
Regional Fine Tuning (Retraining) - Cons
A
- Challenging when comparing experiment results across Regions
- Adds overhead to quota and compute mgmt
