LLM System Design Interview, issue 2, Nov 5, 2025

The Lossless Speedup Trick

AI Engineer interview at OpenAI, and the interviewer asks:

The product team wants a 2x speedup on our Llama 3 70B endpoint, but they’ve forbidden any lossy techniques like quantization or pruning. How can you losslessly accelerate inference, and what core asymmetry in the Transformer are you exploiting?

Don’t say: Well, we could improve our batching strategy with vLLM’s PagedAttention...

How top AI teams double inference speed without quantization, pruning, or touching model weights - by exploiting a core asymmetry in Transformers.

The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.

Read it on Substack

Get the next one

Free on Substack. Unsubscribe in one click.