ML systems might be an area I may explore in the medium term. I am really enjoying this shit.
Ml systems is about llm optimization stuff, writing kernels in cuda, triton etc. from what I know as of now.
Resources:
Find from here - top 100 open source repos -> https://huyenchip.com/llama-police.html
Discord servers like CUDA Mode, tinygrad
repos like VLLM, tensorRTLLM
CUDA systems
PMPP book (I am not interested in books)
Tinygrad codebase -> xjdr suggests to read each line and then learn jax.
Torch / ao