Aditya Singh
Aditya Singh
Thanks a lot for this implementation. I was wondering how can I use the repo to reproduce the results on CIFAR as reported in the paper. As I understand, the...
Corresponding issue: https://github.com/patrick-kidger/equinox/issues/175 - Added `use_ceil` parameter to `Pool` - Added a check to make sure garbage values are not pooled if `padding > kernel/2`.
Corresponds to discussion : https://github.com/patrick-kidger/equinox/issues/136 - Adds `_ordered_tree_map` - Some doc fixes to highlight cases of deserialisation.
Hi, Is it possible to support torch's `ceil_mode` like functionality in Pooling? There is a possible* one sided padding for each dimension as a result of using `ceil` instead of...
Hi, I wanted to check if it is possible for the two operations to produce identical results. I found that `jnp.isclose` assertion fails on the output when comparing `torch` vs...
Hi, Is there a recommended way to partially load weights to a new model ? At the moment I have an inept solution which uses the fact that `fc` attribute...
Hi, Most of the `nn.Modules` ([MLP](https://github.com/patrick-kidger/equinox/blob/a89d5b486d13588caffc095f172a2ec39fd68278/equinox/nn/composed.py#L26)) use `static_fields` for, well seemingly static attributes. In the documentation it is stated that `static_field` should be used rarely. Is there a do's and...
Hi @patrick-kidger , I was wondering if the following modifications are worth adding to `nn.Sequential` 1. Supporting OrderedDicts. They have an ordering but also make it easier to access specific...
Hi, Thank you for sharing the code and model weights for your work. Would it be possible to clarify if the shared weights correspond to the EMA model or not...
Adding a sample training notebook (preferably on each network) on Imagenette demonstrating some different techniques. For example, different optimizers, transfer learning, distillation etc. Open to discussions! - ~~Image Classification~~ -...