Skip to content

Conversation

mikerabat
Copy link

Dear Joapaulo!

I took the liberty to enhance your library with Delphi AVX support and other userfull stuff like faster loading of
classifiers + a few additional checks to avoid AV's . I hope you find these useful and it would be great if you could
take a look at them.

joaopauloschuler#85: Fixed opencl problem

+ Better cloning (way faster)
+ Fixed many Delphi warnings
+ New convolutional classes with swift and gelu
+ Automatic check if AVX is available
+ Fixed issues with output procedures: sometimes error proc pointed to the console output (althouth I wanted that in a file...)
-> an azure VM with 120 cores could only access 60 cores
…r each epoch and after each step

-> the newly introduced method just checks for the callback routines.
Fixed warnings
…ut if the number of examples seen exceeds 2e9 (1 shl 31) examples
-> some changes are still there:
* removed unnecessary extra field formatsettings in TVolume
  -> made a global variable (memory spare ;) )
* fixed many compile warnings and messages. D2010 does not report any
* Updated SelfTest -> avx tests added
* Updated code to reference
* Removed format settings from constantly beeing created in a base class to one single local instance in neuralvolume
-> integrated dynamically into the framework
-> fixed avx512 check in CPUFeatures.pas
-> fixed some compiling problems with FPC
# Conflicts:
#	examples/SelfTest/SelfTest.dpr
#	examples/SimpleFashionMNIST/SimpleFashionMNIST.lpr
#	neural/NeuralAVX.pas
#	neural/NeuralAVXx64.pas
#	neural/neuraldatasets.pas
#	neural/neuralfit.pas
#	neural/neuralnetwork.pas
#	neural/neuralthread.pas
#	neural/neuralvolume.pas
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant