bitanath

Motivated to work on this! Here's my preliminary very basic analysis and approach: I'd like to propose a method with signature `load_state_dict(state_dict:Record)` added in `layers.ts` order to achieve this. Basically...

Hi @Blaizzy is this merged? For inference on a weight normed Conv2D layer isn't it as simple as the below?: ```python weight_v_norm = mx.sqrt(mx.sum(self.weight_v ** 2, axis=(1, 2, 3), keepdims=True))...

So basically just a hook in a custom layer? Can we just simply override Conv2D to include this? Would it also backprop correctly? I have tested the inference on one...

Thanks for your response @awni ! MLX is quite clean and easy to implement models in. Definitely hoping for a torch hub like model zoo with common imagenet pretrained backbones...

To rephrase: Are there any plans to add an MLX pipeline along with commonly supported Keras and PyTorch pipelines to coremltools? This would be extremely useful to have as an...

I'm working on a v2 with a lighter style with the model arch etc available generally: https://github.com/bitanath/ArcaneGANv2