Support for AI Accelerator Cards

[PeterBurner]

Is your feature request related to a problem? Please describe.
Graphics cards are expensive and consume a lot of power. AI accelerator cards (M.2 or PCIe) on the other hand are often cheaper and much more power efficient. There is even an official Raspberry PI AI HAT now.

Describe the solution you'd like
It would be nice if Edit Mind could support at least some of the accelerator cards on the market.
From my research the Raspberry PI AI HAT+ and the Radxa AICore AX-M1 seem to be the most affordable.

Describe alternatives you've considered
Alternatively being able to run the indexing time-boxed on a different host with a dedicated GPU would be an option (like Immichs remote ML).

Additional context
Popular accelerator chips:

• Hailo-8/Hailo-10 --> Raspberry PI AI HAT+, ASUS UGen300, Waveshare Hailo-8 M.2 AI Accelerator Module, Firefly Hailo-8 M.2 AI Acceleration Module
• Axera AX8850 --> Radxa AICore AX-M1, M5Stack LLM-8850, M5Stack AI Pyramid, sipeed Maix4-HAT
• MemryX MX3 --> MX3 M.2 AI Accelerator
• DeepX DX-M1 --> DeepX DX-M1 M.2 Module, Radxa AICore DX-M1
• Metis AIPU --> Metis m.2 AI Acceleration Card

Accelerator integration in other projects:

• Immich ML with Hailo-8
• Immich ML with Axera AX8850
• Frigate Detection with Hailo-8
• Frigate Detection with Axera AX8850
• Frigate Detection with MemryX MX3

[IliasHad]

Alternatively being able to run the indexing time-boxed on a different host with a dedicated GPU would be an option (like Immich's remote ML).

We have support for that, but the remote ML server should have a copy of the video files to process the video and send them back to the local background jobs service.

Would you like to work on adding support for these AI cards?

[PeterBurner]

We have support for [...] remote ML server

Looking at the compose file in the repo again, I can see how this could work 🤔
A bit more documentation would be nice.

Would you like to work on adding support for these AI cards?

I am not familiar enough with Python to attempt that. Sorry.