Open sjfleming opened 2 months ago
Look into torch.compile again
Figure out some way to make the model as fast as we reasonably can.
This is not currently a blocker, but if we don't pursue it, somebody else will come out with a faster version of this.
Nested for-loops in the non-negative least squares solver update are the target.
See Nimish and Ebtisam's example here
https://github.com/cellarium-ai/cellarium-ml/blob/e8a00eeca99c28669ff9174fd66d46ec4422ea14/cellarium/ml/models/custom_logistic_regression.py#L174
Look into torch.compile again
Figure out some way to make the model as fast as we reasonably can.
This is not currently a blocker, but if we don't pursue it, somebody else will come out with a faster version of this.