Feature: Intel ARC GPU support with IPEX

[Disty0]

Pull request checklist

• [*] The PR has a proper title. Use Semantic Commit Messages. (No more branch-name title please)
• [*] Make sure you are requesting the right branch.
• [*] Make sure this is ready to be merged into the relevant branch. Please don't create a PR and let it hang for a few days.
• [*] Ensure all tests are passing.
• [*] Ensure linting is passing.

PR type

• new feature

Description

• Adds Intel ARC GPU support with IPEX on Linux / WSL.

• Patches GradScaler to work with Intel GPU and presents IPEX as CUDA.

• Added some workarounds for no FP64 and DDP support on the main pipeline.

• Intel didn't release a torchaudio wheel so we had to built it from source and requirements-ipex.txt downloads that build (Python 3.10 only).

Screenshot

Localhost url to test on

• http://localhost:7865

[Tps-F]

Nice PR!
Seems to be mostly OK.
I'll fix a few code that bother me later.

[Tps-F]

@Disty0 You hijacked the torch module so have you tried infer?

[Disty0]

These are confirmed to be working on my end:
Infer / Real Time Infer / Voice Splitting / Training

And this video i made as a test (and as a joke) took 25 minutes in total to make.
22 mins to train (a minute per epoch), 30 seconds to split vocal, 30 seconds to infer (takes 5 seconds after the first infer) and another 2 minute to manually combine vocal and instrumental.
(Used a random 10 mins of Jordan Peterson motivational talking video to train.)
https://www.youtube.com/watch?v=FLzHa96SEWQ

And real time infer:
https://www.youtube.com/watch?v=9wIwhPnzgg4

Latency at default settings (+/- 5):
PM: 85 ms
Harvest: 100 ms
Crepe: 110 ms
RMVPE: 130 ms

[fumiama]

Good job with just some small issues.