aimet icon indicating copy to clipboard operation
aimet copied to clipboard

Published 20 hours ago • verified publisher icon quic

AimetTensorQuantizer error

Open zkdfbb opened this issue 1 year ago • 5 comments

We are having the same issue while compiled from source code with python3.10

AimetTensorQuantizer argument error during testing example code

how to solve this problem?

zkdfbb avatar Sep 19 '23 03:09 zkdfbb

@quic-akhobare, could you help answer this?

quic-mangal avatar Sep 19 '23 16:09 quic-mangal

Hi @zkdfbb - This problem happens due to the mismatch in pybind, libpython versions from what I remember. To give you a clue of how this was resolved before.

What version of PyTorch are you compiling against?

quic-akhobare avatar Sep 20 '23 00:09 quic-akhobare

Hi @zkdfbb - This problem happens due to the mismatch in pybind, libpython versions from what I remember. To give you a clue of how this was resolved before.

What version of PyTorch are you compiling against?

Hi@quic-akhobare , the pytorch version is 2.0.0, based on https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-05.html

zkdfbb avatar Sep 21 '23 03:09 zkdfbb

Hi @zkdfbb - This problem happens due to the mismatch in pybind, libpython versions from what I remember. To give you a clue of how this was resolved before. What version of PyTorch are you compiling against?

Hi@quic-akhobare , the pytorch version is 2.0.0, based on https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-05.html

Apologies for the late response. We have not yet tested AIMET against PyTorch 2.0 - so kinda hard to guide you specifically. But I believe this likely boils down to a version mismatch in pybind or libpython..

Possible for you to use PyTorch 1.13 for the short-term because we know that works against AIMET? The AIMET team should start looking at adding/testing support for PyTorch 2.0 but I don't have a firm ETA yet.

quic-akhobare avatar Sep 25 '23 00:09 quic-akhobare

Hi @zkdfbb - This problem happens due to the mismatch in pybind, libpython versions from what I remember. To give you a clue of how this was resolved before. What version of PyTorch are you compiling against?

Hi@quic-akhobare , the pytorch version is 2.0.0, based on https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-05.html

Apologies for the late response. We have not yet tested AIMET against PyTorch 2.0 - so kinda hard to guide you specifically. But I believe this likely boils down to a version mismatch in pybind or libpython..

Possible for you to use PyTorch 1.13 for the short-term because we know that works against AIMET? The AIMET team should start looking at adding/testing support for PyTorch 2.0 but I don't have a firm ETA yet.

Thanks for your advise, I will try and waiting for your testing result.

zkdfbb avatar Sep 26 '23 02:09 zkdfbb