fix: do not share session state across requests in m_serve examples

[markstur]

Examples are holding the session across requests so that different client would share history. This doesn't significantly affect the examples, but it's a bad pattern if the example influences real implementations.

Assisted-by: IBM Bob

Pull Request

Issue

Fixes #1444

Description

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code was added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

Attribution

• AI coding assistants used

---

Adding a new component, requirement, sampling strategy, or tool?

If your PR adds or modifies one of the types below, check the matching box. A checklist of type-specific review items will be posted as a comment.

• Component
• Requirement
• Sampling Strategy
• Tool

NOTE: Please ensure you have an issue that has been acknowledged by a core contributor and routed you to open a pull request against this repository. Otherwise, please open an issue before continuing with this pull request.

[planetf1]

One thing outside this diff's scope, worth a quick follow-up: docs/examples/m_serve/README.md's "Basic Pattern" snippet (line ~115) still shows session = start_session() at module level as the recommended pattern:

from mellea import start_session
...
session = start_session()

def serve(input: list[ChatMessage], ...):
    ...
    result = session.instruct(...)
    return result

That's the exact shared-session pattern this PR is removing from the real examples. Anyone using this README snippet as a starting point for a new example would reintroduce the #1444 bug. Not a blocker for this PR, but worth updating the snippet (move session = start_session() inside serve()) in a follow-up so the docs match the pattern the examples now demonstrate.

Happy to approve once that's done.

[markstur]

One thing outside this diff's scope, worth a quick follow-up: docs/examples/m_serve/README.md's "Basic Pattern" snippet (line ~115) still shows session = start_session() at module level as the recommended pattern:

from mellea import start_session
...
session = start_session()

def serve(input: list[ChatMessage], ...):
    ...
    result = session.instruct(...)
    return result

That's the exact shared-session pattern this PR is removing from the real examples. Anyone using this README snippet as a starting point for a new example would reintroduce the #1444 bug. Not a blocker for this PR, but worth updating the snippet (move session = start_session() inside serve()) in a follow-up so the docs match the pattern the examples now demonstrate.

Happy to approve once that's done.

Done. Thanks for catching that. I feel like I looked at that, but just didn't see it.

[AngeloDanducci]

Should probably update this since the model was updated to
hf.co/ibm-granite/granite-vision-4.1-4b-GGUF:Q4_K_M

[markstur]

Thank you!
fixed