Hanlin Tang

Thanks, we've noticed this internally as well.

Layer buffers in neon are pre-allocated during model creation for a particular batch size. To run prediction on `samples < be.bsz`, you could either regenerate the model with a new...

I am not familiar with spyder IDE, but likely there is a virtualenv setting needed somewhere (or sypder need to be installed in and run from your virtualenv). These may...

Hello, You will first want to use the `MergeBroadcast` container (see `examples/image_caption.py` for an example where you have two inputs) to build your model. For the data side, you can...

We don't have an automatic way of converting python train scripts to yaml. Also note that yaml support can be limited especially for more complex models with aeon-based dataloaders. Is...

Yes I would love to move away from `core`! A few other suggestions/thoughts: * turn folders -> _.py files whenever we don't actually need them * should we be putting...

The fast-fowarding behavior is required for CV, since it needs to replay the random augmentations to set the RNG state properly. We could add in a specific shortcut for NLP...

Thanks @alexriedel1 , for suggesting these. Yes, for the loss functions its just adding them to `loss.py`. We can take care of algorithm-izing them a bit later. For `SnapMix`, see...

@alexriedel1 , just an update -- we implemented adaptive gradient clipping in #924, give it a try/review and let us know if it helps!

Yes, we should add canonical settings for NLP for each algorithm!