TorchSharp
TorchSharp copied to clipboard
dotnet
implementing fake_quantize_per_channel_affine
this PR contains the following changes:
- implementation of
torch.fake_quantize_per_channel_affine - cleanup of
AdjustGammaclass
Error in all builds:
test\TorchSharpTest\TestTorchTensor.cs(7771,19): error CS0623: Array initializers can only be used in a variable or field initializer.
@NiklasGustafsson the Array initializer should have been new double[,,] instead of new double[][][]. have fixed it in a new commit.
Remark: I should build and run tests before committing. However, my local build still doesn't work all the time, so using the CI is my workaround for now 😖
Remark: I should build and run tests before committing. However, my local build still doesn't work all the time, so using the CI is my workaround for now 😖
Understood. Here are test failures:
Failed TorchSharp.TestTensor.TestFakeQuantizePerChannelAffine [22 ms]
Error Message:
System.Runtime.InteropServices.ExternalException : !needs_dynamic_casting<func_t>::check(iter) INTERNAL ASSERT FAILED at "../aten/src/ATen/native/cpu/Loops.h":308, please report a bug to PyTorch.
Exception raised from cpu_kernel at ../aten/src/ATen/native/cpu/Loops.h:308 (most recent call first):
@MovGP0 -- ready to merge? If so, please add something to the top of the RELEASENOTES.md file explaining the API additions. We're working on 0.99.5 now -- planning to release it before the end of the week.
@MovGP0 -- Ping! Is this ready to merge? It'd be great to get a little one-line blurb on this change for the release notes.
Sorry, I had to make a release of 0.99.5 today.
Hey, sorry that I reply this late. Have the championships this and next month and also a new job, so spending all my free time with training or learning 😔 Will probably find time in 4 weeks for further contributions 😊