Tyler Hilbert

Thanks for sharing.
It makes sense the constrained decoding would create the best results for generating valid json.
I was wondering if you have any thoughts on using constrained decoding for generating code for compiled languages, such as C++. Could you utilize a compiler as feedback for the constrained decoder, and if so, would the generated code likely be correct?

What AI advancements do you anticipate for 2026?

This is a great tutorial, I particularly like your 3 TPU optimizations.
What other TPU optimization techniques are possible, beyond SPMD Initialization, FSDP and padding to prevent graph recompilation, if you want to further squeeze better performance out of a TPU?