fix: model options with intrinsics

[jakelorocco]

Misc PR

Type of PR

• Bug Fix
• New Feature
• Documentation
• Other

Description

• Link to Issue: Fixes Intrinsics should default to parameter values in the intrinsic's io.yaml file #954

Verified that io.yaml parameters are used. Verified that model options override these. Added support for all model options with specific caveats for one of cases.

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code as 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

[github-actions]

The PR description has been updated. Please fill out the template for your PR to be reviewed.

[jakelorocco]

Test failures caused by known failure: #970

[jakelorocco]

Intrinsics now support tools. Added tests to ensure that tools are actually getting sent to the model when generating and that generation still works. Required changes to the underlying granite library.

[nrfulton]

FYI @frreiss . Please let us know if there was a reason for granite-io dropped tools. (Fred: we're going ahead with merging this PR but even after this is merged and closed any "there be dragons ahead" warnings would be appreciated so we can get out ahead of them :))

[frreiss]

None of the training data for any of the intrinsics in granitelib-rag includes tool calls. Not sure about other models. In general, if tool calls are not in the model’s training data and the model is fed an input that contains a tool call, then the model is operating outside of the domain it was trained for. A model operating out of domain will produce lower quality results than one operating in domain. The drop in quality is unpredictable, ranging from “we got lucky and it just works” to “it fails completely”. Predicting the level of quality loss a priori is an open research problem.