AI-300 Practice Question 53: Data Asset RO Mount
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 6 · question 3 of 10
A training job reads a 400 GB image folder registered as a uri_folder data asset. The job YAML sets the input mode to download. Copying the folder to the compute node takes over an hour before training starts and nearly fills the local disk. The script only reads the files. Which change removes the copy step?
Show answer and explanation
Answer:
For a uri_folder input, download copies all data to the node before the script runs, while ro_mount mounts the storage location read-only and fetches files as the script opens them. A script that only reads files, and does not need all 400 GB at once, benefits from ro_mount (option D).
- A is wrong because the whole folder would still be copied before training starts, so the long delay remains. A larger disk only avoids running out of space. - B is wrong because in direct mode the job receives the storage URI as a string. No local path is provided, so the script would have to read from storage itself. - C is wrong because a uri_file asset points to a single file and cannot represent a folder of images. Copying files one by one adds the same transfer time.
Reference: Access data in a job