TikTok · Onsite
In processTikTok AI Infra Intern Interviews, 3 Rounds: 1 Experience
A TikTok AI Infra intern candidate described three April 2026 rounds: CPU caches and C++, sparse attention and GPU memory, and graph and stack coding.
At a glance
- Process: three interview rounds, each with a research deep dive, fundamentals and one coding problem (report 01).
- Asked recently: CPU cache lines, page tables and C++ atomics (round 1); sparse attention kernels, distributed-training parallelism and GPU memory reduction (round 2); vtables and
vectormemory (round 3); board connectivity, topological sort, a min stack and Next Permutation (report 01, Apr 2026).
Coverage: 1 report · AI Infrastructure Engineer intern · three onsite rounds · April 2026 · result pending when posted.
What came up
The candidate's Efficient Transformers research anchored every round; the fundamentals swung between computer architecture, C++ internals and large-model training.
Questions we found
Algorithms / Data Structures
| Question | Reports | Last seen |
|---|---|---|
| Board connectivity: is there a path from the first row to the last? | 1 | Apr 2026 |
Topological sort on a given Graph class |
1 | Apr 2026 |
| Stack with O(1) current minimum | 1 | Apr 2026 |
| Next Permutation (LeetCode 31) | 1 | Apr 2026 |
Systems and C++ fundamentals
| Question | Reports | Last seen |
|---|---|---|
| The three parts of a CPU cache line; why multiple cache levels; what happens on a miss | 1 | Apr 2026 |
| Page table overhead and how to reduce it; benefits of virtual memory | 1 | Apr 2026 |
| How C++ function overloading is implemented; C++ atomics | 1 | Apr 2026 |
| When vtable addresses are fixed and when the vptr is set | 1 | Apr 2026 |
How to release the memory a vector allocated |
1 | Apr 2026 |
ML systems
| Question | Reports | Last seen |
|---|---|---|
| Implement a sparse attention kernel in Python | 1 | Apr 2026 |
| Parallelism strategies for training large models | 1 | Apr 2026 |
| Reducing GPU memory in training and inference | 1 | Apr 2026 |
| Writing GPU parallel code | 1 | Apr 2026 |
| How well efficient attention, MoE and distributed training work in production | 1 | Apr 2026 |
1 detailed candidate timeline
01: three rounds anchored on Efficient Transformers research
Interview date: April 2026 · Role: AI Infrastructure Engineer intern · Result: In process
Round 1
- The candidate's research on Efficient Transformers and its main contributions.
- The three main parts of a CPU cache line and what each does; why CPUs use several cache levels; what happens in hardware on a cache miss.
- The overhead page tables add and how to reduce it; the benefits of page tables and virtual memory.
- How C++ function overloading works and how the compiler implements it; what C++ atomic operations are and why they are needed.
- Coding: given a board, decide whether a connected path runs from the first row to the last.
Round 2
- A deeper dive into the Efficient Transformers research: approach, implementation and findings.
- How to implement a sparse attention kernel in Python; parallelism strategies for distributed training of large models.
- How to reduce high GPU memory use in training and inference: sparsification, gradient checkpointing, quantization, mixed precision and knowledge distillation; how to write GPU parallel code.
- Coding: topological sort on a given complete
Graphclass.
Round 3
- The key innovations of the proposed sparse attention mechanism; how effective efficient attention, MoE and distributed training are in real production systems.
- C++ polymorphism: when vtable function addresses are determined and when an object's vptr is initialized; how to release the memory a
vectorholds. - Design a stack with all standard operations plus the current minimum in O(1).
- Coding: Next Permutation (LeetCode 31).
Common questions
How many rounds is the TikTok AI Infra intern interview?
Three in this report, each with research, fundamentals and one coding problem (report 01).
What coding does TikTok ask AI Infra interns?
Board connectivity, topological sort, a min stack and Next Permutation in this report (report 01).
Keep practicing
- TikTok question banks: algorithms, behavioral (sign-in needed).
- Everything TikTok in one place: the TikTok interview page and TikTok hiring insights.
- More firsthand reports: the Experience feed.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.