Prevent Duplicate Model Loading for Voice Cloning TTS

[TheRealPSV]

Is your feature request related to a problem? Please describe.

When using TTS, specifically Qwen3 TTS with voice cloning, I have to create a separate model config file for each individual voice I want to use. This means if I want to use 2 or more different cloned voices at once, the same Qwen3-TTS model file is being loaded on to my GPU in duplicate.

Describe the solution you'd like

Either one of the following:

• An option in a model's config that allows me to specify the relevant parameters for each voice I want to use per-voice within the same model.
• A way (probably also in config) to prevent LocalAI from loading the same model file twice for different model "configs". For instance, if two model config files have parameters.model set to the same value (such as Qwen/Qwen3-TTS-12Hz-1.7B-Base), then reuse the existing version of the model file already loaded into VRAM.

Describe alternatives you've considered

I think the second option above might be the best solution. I can't think of any other alternatives, but it's possible I'm missing something.

Additional context

N/A