Finalise DANRA notebook: merge-artifact fixes + polish (#577)

.github/workflows/install-and-test.yml

@@ -75,24 +75,12 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-meps-reduced-example-data-v0.3.0
 
-      - name: Run smoke test
-        if: matrix.package_manager != 'uv'
-        run: |
-          python -c "import neural_lam; print('neural-lam imported successfully')"
-
       - name: Run tests (excluding notebooks)
-        if: matrix.package_manager == 'uv'
-      - name: Run tests
         run: |
           pytest -vv -s --doctest-modules --ignore=docs/notebooks
 
-      - name: Run notebook tests
-        if: |
-          matrix.package_manager == 'uv' &&
-          (github.event_name == 'push' ||
-           contains(github.event.pull_request.labels.*.name, 'run-notebooks'))
-        run: |
-          pytest -vv -s --nbmake --nbmake-timeout=600 docs/notebooks/
+      # Notebooks run in their own path-filtered, CPU-only workflow
+      # (.github/workflows/notebook-tests.yml).
 
       - name: Upload test figures
         uses: actions/upload-artifact@v4

.github/workflows/notebook-tests.yml

@@ -0,0 +1,42 @@
+# Runs the docs/notebooks/ tutorials end-to-end with nbmake so that API drift
+# in neural_lam breaks the notebook loudly instead of letting it rot. CPU-only.
+# Scoped to PRs that touch code the notebooks depend on (plus every push to
+# main and manual dispatch), so unrelated PRs don't pay the ~3 min cost.
+name: Notebook tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    paths:
+      - "neural_lam/**"
+      - "docs/notebooks/**"
+      - "tests/datastore_examples/**"
+      - "pyproject.toml"
+      - "uv.lock"
+      - ".github/workflows/notebook-tests.yml"
+  workflow_dispatch:
+
+jobs:
+  notebooks:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python 3.13
+        uses: actions/setup-python@v6
+        with:
+          python-version: 3.13
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+
+      - name: Install with uv
+        run: |
+          uv sync --extra cpu --group dev --locked
+          echo "$PWD/.venv/bin" >> $GITHUB_PATH
+
+      - name: Run notebook tests
+        run: |
+          pytest -vv -s --nbmake --nbmake-timeout=600 docs/notebooks/

CHANGELOG.md

@@ -8,6 +8,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [unreleased](https://github.com/mllam/neural-lam/compare/v0.6.0...HEAD)
 
 ### Added
+- Add `hello_world_danra.ipynb` end-to-end tutorial notebook for training on DANRA, with a dedicated path-filtered CI workflow that runs it end-to-end via `nbmake` [\#577](https://github.com/mllam/neural-lam/pull/577) @Sharkyii
+
 - Add `--num_sanity_val_steps` CLI argument to control sanity validation steps before training (#694)
 
 - Add `--train_steps_to_log` CLI option to log training loss for individual unroll steps, and deduplicate common prediction and loss computation steps across loops [\#674](https://github.com/mllam/neural-lam/issues/674) @GiGiKoneti
@@ -48,6 +50,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- `train_model.main` now re-raises on error (`@logger.catch(reraise=True)`) instead of logging and exiting 0, so training/eval failures surface to callers and CI (e.g. the notebook `nbmake` run) rather than passing silently [\#577](https://github.com/mllam/neural-lam/pull/577)
+
+- Allow `graph_lam` training and checkpoint reloads to accept the full set of
+  GNN type CLI options without passing hierarchical-only options to unsupported
+  constructors via a shared `build_predictor` helper ([#686](https://github.com/mllam/neural-lam/issues/686)).
+
 - Fix `RuntimeError` in `HiLAMParallel` forward pass on hierarchical graphs by offsetting edge indices into the global mesh node index space ([#679](https://github.com/mllam/neural-lam/issues/679))
 
 - Fix `IndexError` in HiLAM forward pass by offsetting grid nodes in `zero_index_g2m`/`zero_index_m2g` by the total mesh-node count across all levels ([#642](https://github.com/mllam/neural-lam/issues/642)) @Sir-Sloth-The-Lazy
@@ -197,19 +205,8 @@ This release introduces new features including GIF animation support, wandb run
 
 - Fix Slack domain link [\#288](https://github.com/mllam/neural-lam/pull/288) @sadamov
 
-### Added
-
-- Run notebooks in `docs/notebooks/` as pytest tests via `nbmake` in CI refs [\#69](https://github.com/mllam/neural-lam/issues/69)
-- Add 10-minute timeout for notebook cell execution via `--nbmake-timeout=600` [\#577](https://github.com/mllam/neural-lam/pull/577)
-
-### Changed
-
-- Notebook tests now run selectively: on push to main or when PR has `run-notebooks` label [\#577](https://github.com/mllam/neural-lam/pull/577)
-
 ### Maintenance
 
-- Fix notebook CI failure by adding pytest fixture to create required `danra.datastore.zarr` file and removing duplicate `[build-system]` section in `pyproject.toml` [\#577](https://github.com/mllam/neural-lam/pull/577)
-
 - Update PR template to clarify milestone/roadmap requirement and maintenance changes [\#186](https://github.com/mllam/neural-lam/pull/186) @joeloskarsson
 
 - Update CI/CD to use python 3.13 for testing and full range of current python versions for linting (3.10 - 3.14) [\#173](https://github.com/mllam/neural-lam/pull/173) @observingClouds