Amit Moryossef

Just an update: I wanted to try and take a more data-driven approach: Run every datum through the model, and profile the memory used by the GPU. This means that...

2090Ti: 256x256 resolution > RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 10.76 GiB total capacity; 3.41 GiB already allocated; 9.44 MiB free; 3.46 GiB reserved...

In the config file, I modified every 512 too 256, hoping that would do the trick. Perhaps that was wrong. Now on a V100, with 32GB, 512x512 trains just fine.

I agree with @ttjslbz . @405cddd83a828cec , How can I know if the transcript is aligned to other transcripts?

This is only happening when using angular's `optimization: true` in production which performs minification and tree-shaking. Is the angular/fire compatability code tested under these conditions?

Or any workaround?

For my use case, I would like to perform sign language translation using a sequence of pose vectors (sequence of vectors - to - text)

@Conchylicultor any chance for a review?

Thanks @Conchylicultor I think it is still important to have a custom download method, One extra example I have that I don't see how to adapt would be downloading a...

Hi @Conchylicultor, can you please verify the following steps are correct to have a solution: 1. Add an attribute to the [Resource](https://github.com/tensorflow/datasets/blob/master/tensorflow_datasets/core/download/resource.py#L295) class called `download_fn=None` 2. Send it down the...