feat: register models for llama stack

[feloy]

Signed-off-by: Philippe Martin [email protected]

What does this PR do?

During llama stack container creation:

• all models present on disk are auto-registered on the llama stack

Screenshot / video of UI

(in the demo below, the chat completion with the model MaziyarPanahi/Mistral-7B-Instruct-v0.3.Q4 fails because the output of llama-stack-client models list truncates the names of models - the name to use should have been MaziyarPanahi/Mistral-7B-Instruct-v0.3.Q4_K_M)

You can see in the demo that all models are registered in the llama stack, and inference servers are started only for the ones used during a chat completion.

llama-stack-register-all-models.mp4

What issues does this PR fix or reference?

Fixes #2840

How to test this PR?

[jeffmaury]

Sorry maybe I was not precise enough but we should register all models from AI Lab (if they are not already registered) after the llama-stack container is started without user choice so that user can then play with any model

[feloy]

Sorry maybe I was not precise enough but we should register all models from AI Lab (if they are not already registered) after the llama-stack container is started without user choice so that user can then play with any model

I based my work on what is indicated in the issue:

As a very basic and preliminary work, we would provide a new tab "Llama Stack" with a screen enabling to simply start a llama-stack from Podman AI Lab and choosing a specific model

[feloy]

I'll change this PR to register all models present in AI Lab (models downloaded)

[feloy]

I'll change this PR to register all models present in AI Lab (models downloaded)

Done, ready for review