ngphuoc

Currently, I'd like to change the learning rate. Is creating a new model and copying the old params over for training with the new learning rate the correct approach?

Thanks! That would be simple enough given I am having a callback at every specified epochs/step in my train! function.

I think I also have this GPU memory problem and cannot train a moderate model on a 12Gb GPU. It's a VAE model for the MNIST dataset with 2 MLP...

> Does it also not work on CuArrays#master (using the same memory pool)? If so, please file an issue with a reproducer. It works with CuArrays#master. I then downgrade to...

Yes, I have a similar `LayerNorm` struct to yours. For LSTM we need to apply it to the LSTMCell internal as follow (copied from the Layer Normalization paper): 

`mweWorking'(1.0)` is not working on [email protected] and [email protected], same error!

Thanks. In that case, could you support `create_seq(::Base.Slice{Base.OneTo{Int64}})`? Then mapslices should work!

Thanks, I've added mapslices to the title. Is it likely to be supported in the near future? It seems useful.

Can current ArrayFire.jl do similarly to the following C++ code snippet? ```C++ int n = 8, int m = 8; // dimensions int t = 10; // number of different...

Many thanks!