Felix Schwarz

conda-lock produces lock file which lists a package twice with different versions

5

### Checklist - [X] I added a descriptive title - [X] I searched open reports and couldn't find a duplicate ### What happened? In some situations `conda-lock` produces a lock...

Support pypi repository on the local disk (`file://` urls)

13

This is a quick PR to support `file://` urls which resolves #536 . Since #529 there is support for additional pip repositories. With this PR also local repositories are supported,...

Just by chance: I noticed that the github repo has only two tags. Do you still have some unpushed tags? Pushing these would be most welcome. Otherwise please create a...

hard-coded `WORKING_DIR` in `adaround_weight.py` and `quantsim.py`

2

There is a module-level variable named `WORKING_DIR` in different Python modules. That variable is hard-coded to some path in `/tmp`, e.g. `WORKING_DIR = '/tmp/quantsim/'` ([1](https://github.com/quic/aimet/blob/626913bea8c01339713f5bd1a6372eca2de8fab8/TrainingExtensions/onnx/src/python/aimet_onnx/quantsim.py#L62), [2](https://github.com/quic/aimet/blob/626913bea8c01339713f5bd1a6372eca2de8fab8/TrainingExtensions/onnx/src/python/aimet_onnx/adaround/adaround_weight.py#L67)). Hard-coding the temp directory...

ability to run onnx with quantization nodes outside of aimet?

5

- I have an onnx model which I can quantize using `aimet_onnx.quantsim.QuantizationSimModel`. - aimet seems to create another version of the onnx model in-memory where the new onnx model contains...

Accept a wider range of versions for dependencies

4

Currently when building an aimet wheel there are very strict requirements for dependency versions. I understand that this is likely necessary for the "big ticket items" like pytorch (#2483) and...

After installing the aimet wheel pip complains every time I use it: ``` DEPRECATION: aimet onnx-cpu-1.28.0 has a non-standard version number. pip 23.3 will enforce this behaviour change. A possible...

support for pytorch 2.x?

3

I would like to use aimet directly inside our project which currently uses pytorch 2.0. Is there a timeline on pytorch 2 support?

`--use-cuda` flag is useless - unable to run any code without cuda

2

Currently the code [1](https://github.com/quic/aimet-model-zoo/blob/74731808045ac83542cb899f915917c9bab185ce/aimet_zoo_torch/resnet/evaluator/resnet_quanteval.py#L25) uses ```python parser.add_argument('--use-cuda', help='Use cuda', default=True, type=bool) ``` to parse the CLI parameters. The [documentation](https://github.com/quic/aimet-model-zoo/blob/74731808045ac83542cb899f915917c9bab185ce/aimet_zoo_torch/resnet/ResNet.md?plain=1#L53) mentions that this CLI flag should be used to tell the...

released wheels do not specify dependencies

1

The released wheels do not specify any dependencies even though the code relies on modules outside the standard library. For example [image_net_data_loader.py](https://github.com/quic/aimet-model-zoo/blob/74731808045ac83542cb899f915917c9bab185ce/aimet_zoo_torch/common/utils/image_net_data_loader.py#L17C6-L17C17) requires `torch` and `torchvision`. Also there are dependencies...