[throwaway48476]

It's interesting that they are not separating training and inference.

[noiseinvacuum]

This is specifically designed for inference for recommendations models. It’s not for LLM training or inference.