What is the KV cache, and why can it cause out-of-memory errors when serving an LLM?
Deployment and LLMOps · Pro 2
Short answer
The KV cache stores the key and value vectors for every earlier token, so each new token does not recompute them. It saves work but uses GPU memory that grows with context length, model size and the number of requests served at once, so long chats from many users can exceed memory.
Deeper answer
A rough size is 2 times layers times model width times tokens times batch size times bytes per number. A 13-billion-parameter model with 32 requests of 2,048 tokens can need over 50 GB, more than its weights.
Fixes include shorter contexts and summarised history, serving engines with paged KV memory such as vLLM, models that share keys and values across heads, a quantised cache, and limits on concurrent requests.
What a strong answer covers
- Stores keys and values for earlier tokens
- Avoids recomputation and speeds decode
- Grows with context and batch size
- Names a fix
Answer it out loud in the mock interview about 1 min
Follow-ups to expect
- How would you estimate the cache size for a given model?
- Which serving features help with this?
Common mistakes
- Saying the KV cache is used during training
- Only counting model weights when sizing a GPU
Learn it
Inside LLM inference: prefill, decode and the KV cache
- When would you serve predictions in batch, and when online?
- How would you set up CI/CD for an LLM application?
- What is the difference between data parallelism and model parallelism?
- Why does experiment tracking matter, and what do you record?
- Your model scored well offline but performed badly after launch. What could explain it?
- Explain Kubernetes pods, deployments and services, and why you would use Kubernetes for AI apps.
- How would you load test an LLM feature and set a latency budget?
- Your model provider has an outage during peak hours. What should already be in place?
- How do you manage API keys and secrets across environments?
- How do you detect and handle model drift in production?
- A bank wants your AI system but says no data may leave its network. What are the options?
- Why does an LLM reply have a wait before the first word and then a different speed after it?
- What is quantisation, and what are the trade-offs?
- When would you use a semantic cache for an LLM app, and what can go wrong?
- How would you serve an LLM feature as an API for a web app?
- How would you serve a model that is too big for one GPU, and how would you scale it to more users?
- How would you add single sign-on to an enterprise AI assistant?
- How do you version prompts and models so you can roll back safely?
- What would you capture in traces for an LLM application?
- Why use Docker for an AI application?
- What does an LLM gateway do, and when do you need one?
- Why use vLLM to serve an open-weight model instead of a simple Hugging Face script?