Matthijs Hollemans comments

Results 233 comments of


                                            Matthijs Hollemans

Cannot reproduce the results from the project page

Made a demo for this on Hugging Face Spaces, check it out! 😄 https://huggingface.co/spaces/Matthijs/image2reverb

add new (paid) synthesizer book

Fair enough, I think that is very reasonable. If you want to check out the synth that is being built in the book, the full source code is available here:...

add new (paid) synthesizer book

Thanks for merging! 😄

Question about backprop

Core ML: no, gradients are not exposed in the API. Metal: yes, but you'll have to re-implement the model using the lowest-level MPS primitives, such as MPSCNNConvolutionGradient.

Question about backprop

It's possible a nicer API is available these days (I haven't used MPS in a while) but in the past you had to implement both the forward and backward pass...

Question about backprop

I don't have any examples for this, unfortunately.

Add link to Python libraries for the ANE

That seems to be something different than the Apple Neural Engine?

The accuracy of cpu only, cpu and gpu, and ALL are different, and the result of cpu only is accurate.

They are all accurate, but with different levels of precision. ;-)

Error convert pytorch bert-small-uncased for text classification

@pcuenca Could this be related to the removal of the code that sets the output shape? The error says: ```text Classifier probabilities must have a fully known shape ``` This...

Support OneFormer Model

I'll add it to my to-do list!