Why is temperature-zero decoding not fully deterministic in many APIs?
PICTURE THIS: HOW TO EXPLAIN IT
Simple meaning
Floating-point math, batching, and mixture-of-experts routing can change logits across runs.
WHY — Temperature instead of guessing?
Why interviewers care about Temperature:
on Temperature.
the situation, the default choice, and one exception - that reads as experience.
Name the idea, why it exists, then one short example.
End with when you use it and one common pitfall.
STEPS — What happens step by step?
Before you speak the answer, walk the interviewer through these steps:
- 1Floating-point math, batching, and
mixture-of-experts routing can change logits across runs.
- 2Sparse MoE and load-balancing
noise add extra variance.
- 3If you need reproducibility,
pin model version, seed if offered, and still tolerate rare drift.
- 4Give an example
One tiny concrete case you can say aloud.
- 5Common mistake
What juniors usually get wrong.
- 6Close
When you pick this over the alternative.
EXAMPLE — See it in action
Here's a short line you can speak, broken into clear beats:
Note: Adapt this scaffold to your own project — keep it under 60–90 seconds.
Key takeaway
Floating-point math, batching, and mixture-of-experts routing can change logits across runs. Sparse MoE and load-balancing noise add extra variance.