llama : expose API to retrieve devices associated with the model.

[vlovich]

It's useful from the library to be able to do things like list the features being used by the session. For some reason I'm seeing the model->devices vector be empty so maybe it isn't used anymore?

[vlovich]

@ggerganov is there a process for getting review on a small PR? This has sat around for a little while in limbo so wondering how I can make progress.

[ggerganov]

I am a bit unsure about the usefulness of this API interface. What features are you interested in listing about the session?

[vlovich]

I'm trying to figure out what backends the session is selected to run on.

[slaren]

You can enumerate the available devices using the ggml-backend API. If you want to choose which ones to use, you can do so by setting llama_model_params::devices when loading a model.

[vlovich]

That doesn't really help me though because I'm not trying to tell GGML to use a specific device. I'm trying to figure out which device(s) GGML chose. That's also why the available devices isn't helpful because I don't know what GGML actually ends up picking.