chore: Stack server no longer depends on llama-stack-client

[ashwinb]

This dependency has been bothering folks for a long time (cc @leseb). We really needed it due to "library client" which is primarily used for our tests and is not a part of the Stack server. Anyone who needs to use the library client can certainly install llama-stack-client in their environment to make that work.

Updated the notebook references to install llama-stack-client additionally when setting things up.

[leseb]

I had the extact same branch locally. 😅

You also need to edit pre-commit, it was failing for me:

      - id: provider-codegen
        name: Provider Codegen
        additional_dependencies:
          - uv==0.7.8
        entry: ./scripts/uv-run-with-index.sh run --extra client --group codegen ./scripts/provider_codegen.py

And change the test setup to use --extra client.

[leseb]

Let's add the client as an optional deps

[project.optional-dependencies]
client = [
    "llama-stack-client>=0.3.0",  # Optional for library-only usage
]

[ashwinb]

@leseb I don't see why this is needed. there is nothing which depends on the client, and provider codegen should not need it either.

[ashwinb]

mypy on the other hand needs so I will add it there

[ashwinb]

Also see #4097

[leseb]

How about when we use as a library src/llama_stack/core/library_client.py? This requires both llama-stack and llama-stack-client. So to be it'll be natural in this scenario to do pip install llama-srack[client].

No?

[ashwinb]

But that means the client does get packaged together! That's what we want to avoid.

[leseb]

No, optional dependencies do not automatically get packaged or installed with the main project.
They are only installed if the user explicitly requests them.

wdyt?

I don't consider this a blocker on my end, I'm willing to let it slip if you still strongly disagree. I tread this as a convience / nice to have type of things.

[ashwinb]

In fact optional dependencies DO get packaged. They just don't get installed.

[ashwinb]

One advantage of bundling is that we can bundle the correct version of the client that's supposed to work with this version of the Stack.

[leseb]

In fact optional dependencies DO get packaged. They just don't get installed.

Hum, only metadata is packaged, not the dependency code? I've done some tests

One advantage of bundling is that we can bundle the correct version of the client that's supposed to work with this version of the Stack.

Yes!

[leseb]

Suggested change

	"llama-stack-client is not installed. Please install it with `pip install llama-stack-client`."
	"llama-stack-client is not installed. Please install it with `pip install llama-stack-client` 'uv pip install llama-stack[client]'"

[ashwinb]

I don't see why this is needed, I disagree.

[mergify]

This pull request has merge conflicts that must be resolved before it can be merged. @ashwinb please rebase it. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[leseb]

I don't want to block this work with having llama-stack-client as a optional dep to the project. We can revisit later.

Let's just make pre-commit happy, thanks

[leseb]

@github-actions run precommit

[leseb]

hum has pre-commit bot started its weekend already?

[ashwinb]

@leseb we had to delete the pre-commit bot. It had a security vulnerability. many pull_request_target just have those unless one is extremely careful.

I will update the PR to do what you suggested.

[ashwinb]

The CI failure for docker is due to Huggingface limiting downloads of the nomic model. Landing.