Which asset keeps a training job's dependencies consistent across runs?
Why D is the answer
A registered environment
Reusable environments capture dependencies, so training and deployment match.
In detail
- A registered environment fixes the dependencies for each run.
- Reference the environment by version.
Use case
An illustrative scenario, not a real incident.
A model trains fine in June and fails in July, because a library moved. A registered environment pins the dependencies.
See it in different ways
flowchart LR A["Train"] --> B["Track with MLflow"] B --> C["Register a version"] C --> D["Deploy"] D --> E["Monitor"] E -->|"Threshold exceeded"| A
mindmap
root((Reproducible run))
Code
Git
Data
Versioned data assets
Environment
Registered environment
Compute
Cluster that scales to zero
Results
MLflowflowchart LR S0["Define dependencies"] --> S1["Register environment"] --> S2["Use in each job"]
Key point: Registered environments keep dependencies consistent.
Why the other options are wrong
- A. A storage account tag
- A tag holds no dependencies.
- B. A compute instance schedule
- A schedule starts and stops compute.
- C. A workspace display name
- A display name holds no dependencies.