Update PR templates

.github/PULL_REQUEST_TEMPLATE/pull_request_new-architecture.md

@@ -2,13 +2,39 @@
 
 
 
-# New architecture TODOs
+# Contributor (creator of pull-request) checklist
 
-- [ ] Add your architecture to the experimental folder
+- [ ] Add your architecture to the experimental/stable folder. See the
+  [docs/src/dev-docs/architecture-life-cycle.rst](Architecture life cycle) document for
+  requirements.
   `src/metatensor/models/experimental/<architecture_name>`
 - [ ] Add default hyperparameter file to
-  `src/metatensor/models/cli/conf/architecture/experimental.<architecture_name>.yaml`
+  `src/metatensor/models/experimental/<architecture_name>/default-hypers.yml`
 - [ ] Add a `.yml` file into github workflows `.github/workflow/<architecture_name>.yml`
 - [ ] Architecture dependencies entry in the `optional-dependencies` section in the
   `pyproject.toml`
 - [ ] Tests: torch-scriptability, basic functionality (invariance, fitting, prediction)
+- [ ] Add maintainers as codeowners in [CODEOWNERS](CODEOWNERS)
+
+# Reviewer checklist
+
+## New experimental architectures
+
+- [ ] Capability to fit at least a single quantity and predict it, verified through CI
+   tests.
+- [ ] Compatibility with JIT compilation using `TorchScript
+   <https://pytorch.org/docs/stable/jit.html>`_.
+- [ ] Provision of reasonable default hyperparameters.
+- [ ] A contact person designated as the maintainer.
+- [ ] All external dependencies must be pip-installable. While not required to be on
+   PyPI, a public git repository or another public URL with a repository is acceptable.
+
+
+## New stable architectures
+- [ ] Provision of regression prediction tests with a small (not exported) checkpoint
+  file.
+- [ ] Comprehensive architecture documentation
+- [ ] If an architecture has external dependencies, all must be publicly available on
+  PyPI.
+- [ ] Adherence to the standard output infrastructure of `metatensor-models`, including
+   logging and model save locations.

.github/pull_request_template.md

@@ -0,0 +1,4 @@
+Please go the the `Preview` tab and select the appropriate PR template:
+
+* [Default template](?expand=1&template=pull_request_template.md)
+* [Adding a new architecture](?expand=1&template=pull_request_new-architecture.md)

docs/src/dev-docs/architecture-life-cycle.rst

@@ -16,7 +16,7 @@ removal if maintenance is no longer feasible.
     architecture's internal functionality or any issues that may arise therein.
 
 Experimental Architectures
----------------------------
+--------------------------
 
 New architectures added to the library will initially be classified as experimental.
 These architectures are stored in the ``experimental`` subdirectory within the

docs/src/dev-docs/new-architecture.rst

@@ -46,6 +46,16 @@ In order to follow this, a new architectures has two define two classes
   evaluated and exported. This class must implement the interface documented below in
   :py:class:`TrainerInterface`.
 
+.. note::
+
+    ``metatensor-models`` does not know the types and numbers of targets/datasets that
+    an architecture can handle. As a result, it cannot generate useful error messages
+    when a user attempts to train an architecture with unsupported target and dataset
+    combinations. Therefore, it is the responsibility of the architecture developer to
+    verify if the model and the trainer support the provided train_datasets and
+    validation_datasets passed to the Trainer, as well as the dataset_info passed to the
+    model.
+
 The ``ModelInterface`` is the main model class and must implement a
 ``save_checkpoint()``, ``load_checkpoint()``  as well as a ``restart()`` and
 ``export()`` method.
@@ -57,7 +67,7 @@ The ``ModelInterface`` is the main model class and must implement a
         __supported_devices__ = ["cuda", "cpu"]
         __supported_dtypes__ = [torch.float64, torch.float32]
 
-        def __init__(self, model_hypers, dataset_info: DatasetInfo):
+        def __init__(self, model_hypers: Dict, dataset_info: DatasetInfo):
             self.hypers = model_hypers
             self.dataset_info = dataset_info
 
@@ -135,10 +145,10 @@ these two constants the ``__init__.py`` must contain constants for the original
 
 
 :param __model__: Mapping of the custom ``ModelInterface`` to a general one to be loaded
-    by metatensor-models
+    by ``metatensor-models``.
 :param __trainer__: Same as ``__MODEL_CLASS__`` but the Trainer class.
 :param __authors__: Tuple denoting the original authors with email address and Github
     handle of an architecture. These do not necessary be in charge of maintaining the
-    the architecture
+    the architecture.
 :param __maintainers__: Tuple denoting the current maintainers of the architecture. Uses
     the same style as the ``__authors__`` constant.