Sohaib

- remove protobuf from direct dependencies - replace deprecated mapping.TENSOR_TYPE_TO_NP_TYPE

apple m1 ventura onnx import error

1

protobuf should be removed from direct dependencies for `pip install -e .'[testing]'` to work painlessly on m1 ventura. `pip install onnx~=1.12.0` => builds from source and requires protobuf compiler ```...

add retinanet with resnet backbone

11

- adds resnext to support loading retinanet pretrained on openimages - object detection post processing with numpy. Some parts could easily be written with tinygrad ops but im not sure...

Not sure if there is a clean way of handling the priority of '^R' binding. But I had to explicitly comment out https://github.com/jeffreytse/zsh-vi-mode/blob/master/zsh-vi-mode.zsh#L3471

intel tc

3

**FAR** from ready for review but `OCL_ICD_VENDORS=intel.icd GPU=1 HALF=1 python3 extra/gemm/simple_matmul.py` works - perf is terrible - only fp16 (no integer or bf16)

name conflict with builtin modules

2

When a library uses module names that conflicts with builtins, such as `_ast.py`, the import logic in https://github.com/davidhalter/jedi/blob/master/jedi/inference/compiled/subprocess/functions.py#L139 ends up using `importlib.machinery.BuiltinImporter` as it appears before `importlib.machinery.PathFinder` in `sys.meta_path`.

Thank you for the amazing work. I have tried to use to VGG weights trained on imagenet (https://huggingface.co/brivangl/vgg_kagn11_v2) But the accuracy remains zero so it doesn't seem to be a...