cansik
Onnxruntime 1.14.2
An unofficial fork onnxruntime 1.14.1 with memory leak fixes and build improvements for M1/M2 chips. Up to 1.14.1 (including) builds had a memory leak issue, which was fixed in the next version. But starting from 1.15.0, they have some graphic glitches and library is not suitable at all. https://github.com/s0md3v/roop/issues/321#issuecomment-1585507301
@cansik
PR was re-created due to a typo in a branch name in the previous ones #5
Thank you for the PR.
Merging it into the main branch does not really make sense because the basic functionality of the script is not changed. But I can merge it into a new branch especially for this version, because the main changes have been done in your onnxruntime fork.
I have prebuilt the wheels for v1.14.2, could you please give it a try and tell me if it works? If yes, I can publish them on pypi as well.
@cansik I suggest the community to send you a GIT patch file including the temporary changes, so you can apply it to your build scripts.
Then you can automatically create the official and community version from the upstream repository.
That would be more transparent for everyone as the changes are not hidden in some third party fork.
Your versioning is wrong, you need to add a suffix instead of adding a patch version. Therefore 1.14.1-ce (community edition) would follow semantic versioning as you need to keep align with the upstream.
@henryruhs I have already pre-built the wheels and created a branch for it. No need for sending patch files manually around. What needs to be done is to test the version now by someone who is able to reproduce the bug.
Thank you for the PR.
Merging it into the main branch does not really make sense because the basic functionality of the script is not changed. But I can merge it into a new branch especially for this version, because the main changes have been done in your onnxruntime fork.
That is exactly what I want to ask you.
I have prebuilt the wheels for v1.14.2, could you please give it a try and tell me if it works? If yes, I can publish them on pypi as well.
I've tested "onnxruntime_silicon-1.14.2-cp310-cp310-macosx_12_0_arm64.whl" - everything is good. No memory leaks, no blurred faces. It's the same as my local build.
Thank you.