ML Model Download Progress & Logo

[ax3l]

Simpler version of #433, subset ported over:

• Show download progress from MLFlow
• Add the Genesis AmSC Logo and make it link to MLFlow model catalogue

To do:

• self-review & simplify again
• repeat testing as in [WIP] ML model controls w/ Logos & Progress #433

[EZoni]

Posting here some comments taken out of a Codex review:

1. dashboard/model_manager.py: the AmSC logo image is duplicated once inside a link and once outside it. A tiny helper or single image with conditional link behavior would make this section smaller.

2. dashboard/model_manager.py: AMSC_MLFLOW_MODEL_URL_EXPR duplicates model_type_dict as a Vue ternary. Build the model URL server-side with build_mlflow_model_name() and store it in state, then bind href directly. That removes drift risk and cuts down the expression.

Main advantage: The model URL now has one source of truth.

3. dashboard/app.py + dashboard/model_manager.py: is_model_available_on_mlflow() adds a remote MLflow query, but the "not available" path still calls update(), which constructs ModelManager and attempts the MLflow load anyway. This looks redundant: remove the preflight helper and either always show the download indicator for configs with mlflow.tracking_uri, or keep all MLflow availability/load handling inside ModelManager.

Main advantage: MLflow loading now has one source of truth, ModelManager and the existing load path.

4. dashboard/model_manager.py + dashboard/app.py: the MLflow progress implementation monkey-patches global MLflow classes, while multiple download tasks can be started without cancelling/serializing the previous one. Overlapping downloads can restore each other's patched classes in the wrong order or leave stale progress behavior installed. The more minimal and safer option is probably an indeterminate model_downloading bar only; it would delete the private MLflow imports and most of this context manager.

Main advantage: the dashboard no longer modifies MLflow’s global internals just to show a % progress.

They seem legit improvements to me and could be either separate commits in this PR or separate follow-up PRs. I already tested local patches for each of them, which do make the code a bit more minimal, but will wait to know what others think before going further. We can also discuss them in person if needed - whatever works best.

[ax3l]

I agree with 1-3, feel free to push commits on this PR @EZoni

4: I rather would cancel in-flight downloads. If I wanted an unhelpful, indetermine progress bar I could have avoided a lot of the lifting here. So either ignore or add a cancel.

[ax3l]

For 3 the pre-flight helper is not clear to me. I definitely want to have a is_model_available_on_mlflow() function, otherwise the async code and exception handling gets very messy and error prone.

You want to minimize all exception handling in async functions, because they are notoriously brittle and hard to debug in python.

So do 3. Carefully, it's overly eager in its recommendation.