Open dominicwhite opened 11 months ago
Sorry @dominicwhite for taking so long to look at this issue. I tried running your reproducible example using the dev version of torch and luz and could not reproduce. I feel like this could be related to something like https://github.com/mlverse/torch/issues/961
When I use the
dataset_subset()
function, I get the following error when later training the model with thefit()
function:Reproducible example:
I have found three separate "solutions" that each seem to that fix this issue and allow the model to train without that error:
dataset_subset()
lines.valid_data = valid_dl
argument from thefit()
function.However, I'm not sure why the original code shouldn't work? Why would switching the subset indices (my third solution) fix this?