RepostedTikTok logo

TikTok · Onsite

In process

TikTok AI Infra Intern LLM Round 2: 1 Experience

AI infrastructure engineer · InternshipMaster's degreeApr 2026
Shared by

A TikTok AI Infra intern's second onsite round in April 2026 asked about decoder-only LLMs, Qwen2, PPO versus DPO and RLHF, then Merge K Sorted Lists.

At a glance

  • Process: onsite round 2: LLM fundamentals, training and inference optimization, an internship project, then one coding problem (report 01).
  • Asked recently: why decoder-only dominates, Qwen2 versus Qwen1, PPO and DPO, why RLHF after SFT, and Merge K Sorted Lists (report 01, Apr 2026).

Coverage: 1 report · AI Infrastructure Engineer intern · onsite round 2 · April 2026 · result pending when posted.

What came up

LLM architecture and alignment questions, plus a standard linked-list heap problem.

Questions we found

Algorithms / Data Structures

Question Reports Last seen
Merge K Sorted Lists (LeetCode 23) 1 Apr 2026

LLM fundamentals

Question Reports Last seen
Why is decoder-only the standard LLM architecture? 1 Apr 2026
Qwen2's architecture and its improvements over Qwen1 1 Apr 2026
Main ideas of PPO and DPO 1 Apr 2026
Why RLHF is still needed after SFT 1 Apr 2026
Common ways to optimize training and inference 1 Apr 2026

1 detailed candidate timeline

01: LLM fundamentals and Merge K Sorted Lists

Interview date: April 2026 · Role: AI Infrastructure Engineer intern · Result: In process

Basic knowledge

  1. Why the decoder-only architecture became the standard for most large language models.
  2. What the Qwen2 architecture looks like and how it improves on Qwen1.
  3. The main ideas of PPO and DPO; why RLHF is still needed after supervised fine-tuning (SFT).

Training and inference optimization

  1. Common ways to optimize model training and inference.

Project

  1. An internship project and related research or paper work.

Coding

  1. Merge K Sorted Lists (LeetCode 23).

Common questions

What LLM questions does TikTok ask in AI Infra intern round 2?

Decoder-only design, Qwen2 versus Qwen1, PPO and DPO, and why RLHF follows SFT (report 01).

Keep practicing

Sources: Anywhere useful interview info can be found.

0 total reactions. Your reactions: none.
0 comments

Comments

0

No comments yet — be the first to share what you know.