Parallelize for loop in loss_eval
kristian-georgiev opened this issue · 0 comments
kristian-georgiev commented
The loss_evals can run entirely in parallel. If we vectorize the loops with torch Tensors / numpy arrays, this should speed things up.
kristian-georgiev opened this issue · 0 comments
The loss_evals can run entirely in parallel. If we vectorize the loops with torch Tensors / numpy arrays, this should speed things up.