feat: GeomAI prediction auto-resolution

[msd-11]

Add auto-resolution to GeomAI prediction, setting a resolution is not mandatory and the default value will be automatically selected.

Do not merge until user-api has not been merged.

[sc-38644]
[sc-38520]

[msd-11]

@marielelandais @FlorianJactaAI I've updated the doc and user guide, if you find anything that needs to be updated, please add a comment 👍

[msd-11]

Not sure if we need to remove this line @FlorianJactaAI

[Copilot]

Pull request overview

This PR updates the GeomAI prediction API/documentation to make resolution optional by default (auto-resolved at inference time) and removes explicit resolution usage from user-facing examples.

Changes:

• Updated GeomAIPredictionConfiguration.resolution documentation and removed explicit resolution from the predictions.run(...) example call.
• Adjusted the sampling request payload construction in the GeomAI predictions API client.
• Removed resolution usage/mentions from generative design user guide and examples.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/ansys/simai/core/data/geomai/predictions.py	Updates resolution field docstring and run example to omit explicit resolution.
src/ansys/simai/core/api/geomai/predictions.py	Changes sampling payload to always include resolution.
doc/source/user_guide/generative_design_ug/generative_design.rst	Removes the “Resolution” section from the generative design user guide.
doc/source/examples/02_generative_design_ex/03-interpolate_geometries.py	Removes explicit resolution override from interpolation example.
doc/source/examples/02_generative_design_ex/02-generate_random_geometries.py	Removes explicit resolution override and related tip from random generation example.

Comments suppressed due to low confidence (1)

doc/source/user_guide/generative_design_ug/generative_design.rst:14

• The user guide removed the entire “Resolution” section, but GeomAIPredictionConfiguration still exposes a resolution parameter (now optional/auto-resolved). This leaves users without documentation for (a) how to override the auto-selected resolution, and (b) any remaining limits/performance considerations when manually setting it. Consider replacing the deleted section with an updated explanation of auto-resolution and the manual override behavior instead of removing it entirely.

The latent parameters (``latent_params``) correspond to a list of numbers (floats) that represent the position of the geometry in the latent space.
You define this parameter to generate a geometry with a trained model.

The number of floats must match the ``nb_latent_param`` your model was requested with.
For more information, see Number of latent parameters.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

run_geomai_sampling now always sends a JSON body with {"resolution": None} when resolution isn’t provided. This changes the previous behavior (omitting the key entirely) and may break the endpoint if it distinguishes between a missing field and null. Consider restoring the conditional payload construction (only include resolution when not None) or confirm/update the server contract and add a test that covers the resolution=None case.

Suggested change

	payload = {"resolution": resolution}
	payload = {}
	if resolution is not None:
	payload["resolution"] = resolution

[Copilot]

The resolution field docstring is a multi-line docstring but the closing triple quotes are on the same line as the final sentence, which will fail pydocstyle/Ruff (D209: “Multi-line docstring closing quotes should be on a separate line”). Move the closing """ to its own line (and drop the trailing space before it).

Suggested change

	As a result, the resolution is not fixed per model but varies per inference, offering a data-driven default that improves quality while remaining editable by the user. """
	As a result, the resolution is not fixed per model but varies per inference, offering a data-driven default that improves quality while remaining editable by the user.
	"""

[FlorianJactaAI]

This is nice! I think we want to still mention it somewhere. In the case of Stellantis, they had geometries that were created with decent resolution but because of how thin the geometry was, they had to go to 900.

When working with complex or thin geometries, you might want to increase the resolution to get the best result.

-> something like this?

A side comment: Does the random sampling has been released? If so, I should also change the code so that we don"t do this random sampling I created.

[marielelandais]

Waiting for more info before being able to provide my feedback.

[marielelandais]

I'm not sure we should have removed the entire section, though. I suggested changes for validation in Shortcut.

[marielelandais]

Here as well, I suggested changes for validation in Shortcut.