A municipal animal-shelter log has mixed string and number columns, blank feeding times, and volunteer rows that should be dropped. How should that prep stay on the GPU?
Select an answer to reveal the explanation.
Short Explanation
Mixed types, blank feeding times, volunteer rows to drop. Cast, fill or drop, and filter with cuDF so that prep never leaves the GPU table. Pandas samples, a later fine-tune, and a NIM type-fix are host bounces.
Full Explanation
Essential GPU prep is type repair, missing-value handling, and row filters on cuDF. Mixed types, blank feeding times, and volunteer rows are that work, and they should not bounce to pandas. This is not a later training loop and not a serving fix. Keep the cleaned table on the GPU.