What can I run without paying?

With a free account, any topic at easy difficulty, up to 20 items per test, and every explanation shown. Premium adds the medium and hard tiers, adaptive runs, and tests up to 100 items.

Will a second test show me the same items?

Not until you run out. Your browser remembers which items you answered, and the next test deals unseen ones first. Repeats start only once a topic and difficulty pool is used up. Current bank size: 360 items.

Should I take the assessment or a practice test?

Both, in that order. The assessment checks all 12 topics once and turns the misses into a study plan. A practice test goes narrow: one topic, many items, a score you can compare against your next attempt.

Which topics are in the bank?

The 12 rounds of an inference loop: Inference Physics & Roofline, KV Cache & GPU Memory, Batching, Scheduling & Admission, Attention Kernels, CUDA, Triton & GPU Performance, Quantization & Numerics, Speculative & Constrained Decoding, Distributed & Disaggregated Inference, Serving Engine Internals, Inference System Design, Performance Debugging & Benchmarking, Implement-from-Scratch Coding.

Practice tests · 360 items · 12 topics

Drill the rounds before the panel does.

Choose one topic or all of them, a difficulty and a length. Each item is a short serving scenario: a KV budget to size, a profile to read, a kernel choice to defend. After you answer you get the reasoning and a link to the full worked question. Later tests deal items you have not answered yet, so a second score reflects study rather than recall.

Configure a test

Topic

Difficulty

Questions

120 in this pool · about 7 min

Reveal answers

Two sample items

Taken from the bank with the answer and explanation visible. In a real test the explanation stays hidden until you have picked an option.

KV Cache & GPU Memory · easy · sample

Llama 3.3 70B has 80 layers, 64 query heads, 8 KV heads and head dimension 128. How many bytes of KV cache does one token occupy in BF16?

Reveal the options and the answer →

2,560 KiB, from 2 x 80 x 64 x 128 x 2 B

160 KiB, from 80 x 8 x 128 x 2 B

✓320 KiB, from 2 x 80 x 8 x 128 x 2 B

4 KiB, from 2 x 8 x 128 x 2 B

Per token the cache holds a K vector and a V vector for every KV head in every layer: 2 x 80 x 8 x 128 x 2 B = 327,680 B = 320 KiB. Using the 64 query heads is the classic GQA mistake and inflates the answer 8x; query heads share the 8 KV heads in groups of 8. Dropping the leading 2 counts only K or only V. 4 KiB is one layer's worth. At 8,192 tokens the 320 KiB becomes 2.5 GiB per sequence, the number every capacity plan starts from.

Performance Debugging & Benchmarking · easy · sample

TTFT doubled right after a prompt-template deploy. Request rate, prompt lengths and GPU utilization look unchanged. What is the first suspect?

Reveal the options and the answer →

The new template uses more output tokens, so each request now spends longer in the decode phase

The GPUs started thermal throttling at the same time, which lowered the clocks for every prefill

The tokenizer was upgraded with the template, and the new one is slower on the CPU for every request

✓A per-request value such as a timestamp now sits ahead of the shared prompt, breaking prefix hits

An early per-request token can invalidate that prefix block and every chained descendant. In the synthetic case with 2,560 shared and 440 unique tokens, a 95% shared-token reuse rate gives 440 + 0.05 x 2,560 = 568 computed tokens/request; zero reuse gives 3,000, a 5.282x work ratio. Compare tokenized prompts and cache identity, then matched per-replica counter deltas. The work ratio does not predict a p99 multiplier, and unchanged utilization does not establish unchanged useful work.

12 topics, three tiers

Easy checks definitions and first-order arithmetic. Medium asks you to apply a mechanism to a specific model, GPU and batch. Hard covers second-order effects: the regime change, the regression after an upgrade, the edge case in a kernel. Each topic has its own question set and concept track, so a low score points at what to read next.

How tests fit with the rest of the site

First: the assessment

The free assessment samples all 12 topics and builds a study plan from the ones you missed.

Take the assessment →

Then: repeated drills

Learn mode shows the explanation after each item. Exam mode holds every answer until you submit. Treat the score as a recognition check: picking the right option is easier than deriving a KV budget or debugging a profile with an interviewer watching.

Finally: the worked answers

Each explanation links the concept or question it came from. The library holds 196 questions with full worked answers and 36 concept pages.

Open the question library →

Practice test FAQ

What can I run without paying?: With a free account, any topic at easy difficulty, up to 20 items per test, and every explanation shown. Premium adds the medium and hard tiers, adaptive runs, and tests up to 100 items.
Will a second test show me the same items?: Not until you run out. Your browser remembers which items you answered, and the next test deals unseen ones first. Repeats start only once a topic and difficulty pool is used up. Current bank size: 360 items.
Should I take the assessment or a practice test?: Both, in that order. The assessment checks all 12 topics once and turns the misses into a study plan. A practice test goes narrow: one topic, many items, a score you can compare against your next attempt.
Which topics are in the bank?: The 12 rounds of an inference loop: Inference Physics & Roofline, KV Cache & GPU Memory, Batching, Scheduling & Admission, Attention Kernels, CUDA, Triton & GPU Performance, Quantization & Numerics, Speculative & Constrained Decoding, Distributed & Disaggregated Inference, Serving Engine Internals, Inference System Design, Performance Debugging & Benchmarking, Implement-from-Scratch Coding.