In inference that's often the case, but numerical stability in training is often...

6gvONxR4sf7o 37 days ago | parent | context | favorite | on: X X^t can be faster

In inference that's often the case, but numerical stability in training is often a massive lift for new algorithms.