I look forward to following mirage development.

I was wondering if these same optimizations can be brought to bear on training as well, rather than only inference. I guess the challenge here is fusing backward computations with gradient communication.

I also saw that this currently does not handle dynamic workloads such as MoE. I recently came across this paper that does exactly this:

FlashDMoE: Fast Distributed MoE in a Single Kernel - https://arxiv.org/pdf/2506.04667

I cherish writing and find it a wonderful tool for thinking. So far, I've tried to do technical writing without much LLM help. I do run the final writing through a good model to point out factual inaccuracies.

Much work on generalization of ML models deals with asymptotic bounds. Here, there's a precise way of measuring these, even for relatively large models with millions of parameters.