feat(tabpfn): add TabPFN-3 support (v0.0.33)

[AzulGarza]

Summary

• Add TabPFN-3 support via foundationforecast>=0.1.6 (now on PyPI), keeping TabPFN-2 as the default checkpoint through model_path
• Re-export TABPFN_V2_MODEL / TABPFN_V3_MODEL, update mock conftest, and add LOCAL integration tests for v2/v3 (point, quantiles, level)
• Add TabPFN family notebook, model hub updates, CI TABPFN_TOKEN + checkpoint cache, and v0.0.33 changelog

Test plan

• uv lock --upgrade-package foundationforecast resolves 0.1.6 from PyPI
• pytest tests/models/foundation/test_tabpfn.py -m models (unit tests pass; integration tests skip without TABPFN_TOKEN)
• CI green on all Python versions
• Merge and tag v0.0.33 after review

Made with Cursor

[Copilot]

🔵 Needs a closer look

Two moderate test-coverage issues and an unregistered pytest marker remain.

Pull request overview

Adds TabPFN-3 support in v0.0.33 while retaining TabPFN-2 as the default checkpoint.

Changes:

• Updates dependencies and exposes TabPFN v2/v3 constants.
• Adds mocked, unit, and LOCAL integration tests.
• Updates documentation, examples, changelogs, and CI caching/token configuration.

File summaries

File	Summary
uv.lock	Locks updated forecasting dependencies.
timecopilot/models/foundation/tabpfn.py	Exposes TabPFN v2/v3 model constants.
tests/models/foundation/test_tabpfn.py	Adds TabPFN tests. Moderate findings remain around silently skipping invalid-token V3 tests and hiding supported-version dependency failures; the models marker is also unregistered.
tests/models/conftest.py	Adds a mocked TabPFN predictor.
pyproject.toml	Updates dependencies and package version.
mkdocs.yml	Updates documentation navigation.
docs/model-hub.md	Documents TabPFN versions and licensing.
docs/examples/index.md	Lists the TabPFN notebook.
docs/changelogs/v0.0.33.md	Documents the release changes.
docs/changelogs/index.md	Links the new changelog.
.github/workflows/ci.yaml	Configures token injection and checkpoint caching.

Review details

Suppressed comments (3)

tests/models/foundation/test_tabpfn.py:95

• These V3 integration tests treat any TabPFNLicenseError as a skip even though _require_tabpfn_token() has already confirmed that a token is present. A CI run with an invalid or unaccepted token can therefore skip every V3 case and still pass, so the advertised V3 coverage is not gating; make this configuration error fail CI or provide an explicitly configured opt-out instead of silently skipping it.

    except TabPFNLicenseError as exc:
        if model.model_path == TABPFN_V3_MODEL:
            pytest.skip(
                "TabPFN-3 license not accepted; accept at https://ux.priorlabs.ai"
            )

tests/models/foundation/test_tabpfn.py:10

• tabpfn_time_series is a mandatory dependency of foundationforecast on the supported Python 3.10–3.12 interpreters, so importorskip would silently hide a broken installation and skip this entire test module. Restrict the module-level skip to Python >=3.13 and let missing dependencies fail on supported versions.

pytest.importorskip("tabpfn_time_series", reason="TabPFN requires Python < 3.13")

tests/models/foundation/test_tabpfn.py:24

• The new models mark is not registered in [tool.pytest.ini_options].markers, so the documented pytest ... -m models command emits PytestUnknownMarkWarning. Register the marker in pyproject.toml (or remove the mark) so this test selector is supported by the repository configuration.

pytestmark = pytest.mark.models

• Files reviewed: 10/12 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.