Sasha Rush
Sasha Rush
### Description of the model to be implemented If I have 6 layers stacked of a model (think transformer style), I would like to have their setup be batched by...
Hi there, I really love the g9.js demos! I think they convey how gradients work in such a nice and intuitive manner. Just for fun I was playing with the...
From conversations with @justinchiu The current distributions framework need a bit of an overhaul. We think this is an area with a lot of potential, so we are thinking through...
There seems to be use cases for these two functions, so we should add them. Unlike torch function they should not require (or allow) size 1 dimensions but just add...
Implement step size for slice.
Semantic is not completely correct for copying in indexes yet. Need to add better tests here.
Some folks have suggested we might want to add optional dim labels as well. I was skeptical of this at first, but it might be worthwhile. For instance if you...
@da03 Now that we have ntorch.nn what if we overload some of the parameter functions? class MyMod(ntorch.nn.Module): ... And then internally if you do `self.register_named_parameter("w", ntorch.tensor()) ` it makes a...
It is often convention to have the fixed dimensions in a numpy array be leading dimensions. However it seems like this syntax is not supported. Would it be possible to...