feat: macOS support

[jeethu]

I've verified on my Linux box with NVIDIA gpus that this change is fully backwards compatible. Should work on any Apple Silicon Mac with at least 24GB of RAM or even 16GB if no other apps are running. Some caveats:

1. The warping module cannot be run on MPS, because aten::grid_sampler_3d isn’t yet implemented in PyTorch’s MPS backend. The PYTORCH_ENABLE_MPS_FALLBACK env var can be set to work around it.
2. Also, onnxruntime-gpu cannot be installed on macOS, so I've substituted it with onnxruntime-silicon.
3. --flag_do_torch_compile doesn't work on macOS because the PyTorch MPS backend doesn't support it, yet.

Install requirements with:

pip install -r requirements_apple.txt

Run with:

PYTORCH_ENABLE_MPS_FALLBACK=1 python app.py

[zzzweakman]

Thanks so much for your PR! We tested it on macOS and everything works great.

Before merging into the main branch, we need to make some updates to the README and other formatting changes. We'll create a new branch for this and merge your changes there first. Once everything is ready, we'll merge it all into the main branch and release it as a new feature.

Thanks again for your contribution! @jeethu

[Fronkielein]

Thanks so much @jeethu and @zzzweakman for your efforts. I am an absolute beginner and I desperately tried to run LivePortrait on my Mac M1 until yesterday. I have no idea of how to install the new feature as soon as it is there, but I hope I'll find out. I'm just happy that there is a M1 solution now.

[shaoguowen]

Thanks so much @jeethu and @zzzweakman for your efforts. I am an absolute beginner and I desperately tried to run LivePortrait on my Mac M1 until yesterday. I have no idea of how to install the new feature as soon as it is there, but I hope I'll find out. I'm just happy that there is a M1 solution now.

Hi, What is the speed on the M1?

[Fronkielein]

HI, it took 100 seconds of generation time for 97 frames (which is almost 4 seconds of video) Apple M1 Max with Apple GPU 64 GB RAM Mac Sonoma 14.5 (Mac Studio M1Max/64/2TB 24C GPU BTO)

…

Am 18.07.2024 um 07:57 schrieb shaoguowen ***@***.***>: Thanks so much @jeethu <https://github.com/jeethu> and @zzzweakman <https://github.com/zzzweakman> for your efforts. I am an absolute beginner and I desperately tried to run LivePortrait on my Mac M1 until yesterday. I have no idea of how to install the new feature as soon as it is there, but I hope I'll find out. I'm just happy that there is a M1 solution now. Hi, What is the speed on the M1? — Reply to this email directly, view it on GitHub <#143 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BJZASCBG5T2UDMQM7FKRKDTZM5KOJAVCNFSM6AAAAABK3GFJSGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMZVGY2TMMRXHE>. You are receiving this because you commented.

[wming3]

inter cpu
torch insert failed
#295

[yourmoonlight]

HI, it took 100 seconds of generation time for 97 frames (which is almost 4 seconds of video) Apple M1 Max with Apple GPU 64 GB RAM Mac Sonoma 14.5 (Mac Studio M1Max/64/2TB 24C GPU BTO)
…
Am 18.07.2024 um 07:57 schrieb shaoguowen @.***>: Thanks so much @jeethu https://github.com/jeethu and @zzzweakman https://github.com/zzzweakman for your efforts. I am an absolute beginner and I desperately tried to run LivePortrait on my Mac M1 until yesterday. I have no idea of how to install the new feature as soon as it is there, but I hope I'll find out. I'm just happy that there is a M1 solution now. Hi, What is the speed on the M1? — Reply to this email directly, view it on GitHub <#143 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/BJZASCBG5T2UDMQM7FKRKDTZM5KOJAVCNFSM6AAAAABK3GFJSGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMZVGY2TMMRXHE. You are receiving this because you commented.

Hmmm, This speed is a bit slow, Do we have a way to accelerate on m1max?