alexander-g

I don't know how to solve this but here are some hints: - That line `commandBuffer->copyBuffer` is in the function `Tensor::recordCopyFrom()`. - This function was called most likely from `OpTensorCreate::record()`...

@axsaucedo I've taken a short look at this and have some questions: - What's the point of staging tensors? Especially, why give the user the option to create one manually?...

3D is still not enough for many use cases like CNNs. Consider using something like [xtensor](https://github.com/xtensor-stack/xtensor) in combination with push constants for shape/stride parameters. It supports N dimensions and has...

Bump, I am interested in this. It would be nice if it was possible to sync only a part of a tensor with the device/host. Also, specifying tensor offsets for...

Another option (something I am planning to do) is to write a backend for [Jax](https://github.com/google/jax). Jax re-implements Numpy with additional features like gradient calculations and just-in-time compilation of functions. The...

An update from my side: I've went public with my [vkJAX](https://github.com/alexander-g/vkJAX) project, a JAX interpreter for Vulkan. As of now it only covers an **incomplete** subset of all JAX/XLA primitives...

- This PR is finished but depends on PR #115 so I'm not marking it as ready yet - Includes UNets with ResNet18 and ResNet50 backbones pretrained on COCO with...

No, the data generators can not help with this. The problem is that the training loop is waiting for the callbacks to finish before starting the next training step. In...

> I have one main question: are the parameters being transferred to the new module thanks to Module.set_default_parameters? Parameters are transferred but I do not call `set_default_parameters` explicitly. I construct...

> Interesting. In the latest version I think you have to be more explicit about this but its just a single line of code. You're right, it's required, I've added...