VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
u_Unrolled
All paths

Path · Advanced

Inference Engineering

Serve LLMs fast and cheap: GPUs, the KV cache, batching, quantization, parallelism and cost per token.

31 lessons5 stagesabout 22 hours4 free
Start with “The Transformer I”
You will build: A benchmark for LLM serving

The roadmap

  1. 1

    What you are serving

  2. 2

    Systems underneath

  3. 3

    Part 9: Serve it fast

  4. 4

    GPU engineering and CUDA

  5. 5

    Cloud and platform engineering

Every lesson in this path is written.

Practise what you learned on the AI Engineer Sheet, or build one of the projects.

Explore other paths