This is very narrowly focused on LLMs, whereas triton is still useful for running all kinds of ML models. In practice, Triton is a very poor choice for LLMs spe

This is very narrowly focused on LLMs, whereas triton is still useful for running all kinds of ML models. In practice, Triton is a very poor choice for LLMs specifically because it has none of the required non negotiable features like KV caching built in.