Threading vs multiprocessing vs asyncio: which one for which ML task?

You need to do three jobs:

1. Call an embedding API for 100,000 texts.
2. Clean and tokenise 50 GB of text with pure-Python logic.
3. Keep a GPU busy while the CPU prepares the next batch.

For each job choose threading, multiprocessing or asyncio, and justify it. Then compare the three models in general.