Aakash Kumar Nain

Support for tf.keras

10

As `tf.keras` is now the official TF high level API, are there any plans to support models trained in `tf.keras`?

Confusion Matrix

1

As per the discussions on Kaggle, yours implementation is the only implementation that is fully correct for the given metric but there is one thing that I couldn't understand as...

This extension looks pretty good and will be very useful for many cases. But is it possible to use it directly with simple `jupyter notebook` instead of `jupyter lab`?

MaskedLMHead embeddings

12

I noticed a change that was introduced in the `MaskedLMHead` layer, and it broke my entire workflow. Earlier we had the signature for `MaskedLMHead` like this: ```python out = keras_nlp.layers.MaskedLMHead(...

GPU memory keeps increasing

1

I cloned this repo and tried to evaluate the model on a number of images. For each image evaluation, the GPU memory keeps increasing. How can I control that?

[Enhancement] Adding `dtype` to nn layers

9

Moving the discussion from #673 to here. As discussed, adding `dtype` for creation of layers is necessary as it allows to load modules/layers directly with a specific precision (very useful...

Masking in Embedding layer

1

When the input sequences are of different lengths, padding with zeros is a common thing to make the shape of every sample in a batch of the same size. But...

Recommended way of filtering params for weight decay

8

Apologies if this has been asked before but I couldn't find any example that demonstrates this in a simple manner. I have a model built in Equinox. Now, I want...

Right way to count cache size

2

I am trying to keep a count of how many times the `__call__` method of a pytree is compiled and what's the size of the the cache. For a jitted...

[Question] Best practices for profiling Equinox models

9

I am building GPT kinda model in Equinox, and right now the forward pass is extremely slow compared to my torch implementation. I think this is one of the cases...