How LLMs Work

Understand the model you're calling. Implement tokenisation, attention and positional encodings, the inference tricks that make serving affordable (KV cache, GQA, speculative decoding) and the sampling methods that shape every response.

19 problems · 510 XP · 2 tracks