Sam Ade Jacobs
Sam Ade Jacobs
Thank you the nice tutorial and supporting code. I made a plot (attached) of KL Loss vs iterations of your implementation and that of Keras ([blog](https://blog.keras.io/building-autoencoders-in-keras.html), [code](https://github.com/keras-team/keras/blob/master/examples/variational_autoencoder.py)). Could you please...
LBANN implementation of multi generative model (multi GAN) for 3D NERSC Cosmology application. Subgraph parallelism of multi discriminators network is optional (can be set by enable_subgraph flag) This PR depends...
This PR extends previous implementations of data transformation/augmentation to work in training mode only.
There is a hidden bug in use/implementation of _num_samples_, _num_train_sample_s and _num_test_samples_ flags in data_reader_smiles. For example, if I specify values greater than num of samples in the dataset the...
Whereas LBANN uses keep probability in Dropout layer,[ Pytorch](https://pytorch.org/docs/stable/nn.html#dropout-layers) and [TensorFlow](https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras/layers/Dropout) use dropout probability (rate). I suggest we adopt Pytorch/TF2.0 approach in LBANN.
Unit test implementation for MOE residual matrix multiplication.
I am curious as to the purpose of the SafeInvLayer no where is still explained or document. I know what it does (that is inverse of its input value x...
# What does this PR do? This PR enhances capabilities of [DeepSpeed long sequence (context) parallelism (aka DS Ulysses)](https://dl.acm.org/doi/10.1145/3662158.3662806) with support for HF models. Support is currently enabled when both...