Emile Aydar
Results
2
comments of
Emile Aydar
> We already had a number of problems in the past due to `vmap`, so we decided to add the possibility to simply avoid it by setting `parallel_chunk_size=1` when calling...
> Also, do you think it could happen that people want to differentiate _with respect to_ a sparse tensor (e.g. have a sparse parameter tensor in their model)? In this...