[WIP] Flux feature parity

[MartinuzziFrancesco]

Addressing #12

[codecov]

Codecov Report

Attention: Patch coverage is 97.29730% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 96.52%. Comparing base (273a0cf) to head (2af0c18).

Files	Patch %	Lines
ext/WeightInitializersCUDAExt.jl	93.75%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #13      +/-   ##
==========================================
+ Coverage   95.77%   96.52%   +0.75%     
==========================================
  Files           4        4              
  Lines          71      144      +73     
==========================================
+ Hits           68      139      +71     
- Misses          3        5       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[MartinuzziFrancesco]

I think I could streamline the tests a little, there are sections that can be shortened a bit. Beside that, create_bias is missing, but I think that is more a DL library specific implementation

[avik-pal]

yeah create_bias is a bit specific. Can you rebase, then we should be good?

[MartinuzziFrancesco]

This should be good to go now, I can't see the failed tests on buldkite thou

[avik-pal]

I made buildkite public, but rn the cuda clusters have a disk space issue. Let's merge it once that is resolved

[avik-pal]

this line is not needed, right? Just LinearAlgebra needs to be put inside @recompile_invalidations

[MartinuzziFrancesco]

ah yes, I missed a couple of details in the rebase apparently. I fixed that and also added the missing inits to the non differentiable list