Tokenizers & Model Serving

The tokenizer is one of Perplexity's two recurring 4-hour OAs. The other is the versioned store in the practical problems lesson. You start from a slow reference and write a fast version whose output is identical, inside a time limit, and a later part pushes past what you can compute in full.

Phone screens ask for smaller serving pieces, on the machine learning engineer track and in technical screens, sometimes with the signature and unit tests already written.

None of these need a real model. A table or a stub stands in for one, so every answer can be checked exactly. The tokenizer work is mostly about doing the same thing much faster, and the serving problems are about the code around a model.

PerplexityPremium lesson · Coding round

The 4 Perplexity problems are below.

1. 1A Premium questionCoding · Hard · Frequently asked
2. 2A Premium questionCoding · Medium
3. 3A Premium questionCoding · Medium

+ 1 more in this lesson

Each with hints, the follow-up and a tested solution.

Get the Perplexity questions

$49 a year · all 100 companies · cancel anytime

From 20,000+ candidate reports, forum threads behind logins and write-ups in Chinese.

Read a free Perplexity lesson firstAlready a member? Sign in