ML Coding: Top-p Sampling and Multi-Head Attention in NumPy

Problem Overview

You implement four functions in three graded parts, in NumPy, not PyTorch, in a notebook that already defines softmax and comes with tests you have to pass. Part 3 builds on Part 2: Part 1 — Top-p sampling. top_p_filter(logits, p) keeps the smallest set of most-probable tokens whose total probability…

• 1 candidate-reported follow-up — the exact probes interviewers asked, with the trigger for each
• The rest of the problem statement — full requirements, constraints, and edge cases
• Approach and trade-offs — what passing candidates did, and the mistakes that sink people