LoRA support

Hi,

I use MLX to create LoRA adapters in .safetensors format.

MLX LM has a web server that can take the name/path of the LoRA adapter at inference time when using the Chat completion endpoint by specifying the adapter parameter in the JSON payload.

Would be great to see the same being supported on LM Studio

Start by locating the MLX engine's web server and Chat completion endpoint, then compare its request handling with MLX LM's LoRA adapter parameter. The issue describes .safetensors adapters and an adapter name/path in the JSON payload; done means LM Studio accepts and uses that adapter during inference.

Written by the indexing model from the issue text.