Use singular for property names in output metadata

docs/src/outputs/charges.rst

@@ -3,8 +3,8 @@
 Charges
 ^^^^^^^
 
-Charges are associated with the ``"charges"`` key in the model inputs, and must
-adhere to the following metadata schema:
+Charges are associated with the ``"charges"`` or ``"charges/<variant>"`` name
+(see :ref:`output-variants`), and must have the following metadata:
 
 .. list-table:: Metadata for charges
   :widths: 2 3 7
@@ -36,8 +36,8 @@ adhere to the following metadata schema:
     - the charges must not have any components
 
   * - properties
-    - ``"charges"``
-    - charges must have a single property dimension named ``"charges"``, with a
+    - ``"charge"``
+    - charges must have a single property dimension named ``"charge"``, with a
       single entry set to ``0``.
 
 The following simulation engine can provide ``"charges"`` as inputs to the models.

docs/src/outputs/energy.rst

@@ -3,8 +3,8 @@
 Energy
 ^^^^^^
 
-Energy is associated with the ``"energy"`` or ``"energy/<variant>"`` key (see
-:ref:`output-variants`) in the model outputs, and must have the following metadata:
+Energy is associated with the ``"energy"`` or ``"energy/<variant>"`` name (see
+:ref:`output-variants`), and must have the following metadata:
 
 .. list-table:: Metadata for energy output
   :widths: 2 3 7

docs/src/outputs/features.rst

@@ -15,8 +15,8 @@ by a neural-network or a similar machine learning construct.
 .. _Atom-centered symmetry functions: https://doi.org/10.1063/1.3553717
 
 In metatomic models, they are associated with the ``"features"`` or
-``"features/<variant>"`` key (see :ref:`output-variants`) key in the model
-outputs, and must adhere to the following metadata specification:
+``"features/<variant>"`` name (see :ref:`output-variants`), and must have the
+following metadata:
 
 .. list-table:: Metadata for features output
   :widths: 2 3 7

docs/src/outputs/masses.rst

@@ -3,8 +3,8 @@
 Masses
 ^^^^^^
 
-Masses are associated with the ``"masses"`` key in the model
-inputs, and must adhere to the following metadata schema:
+Masses are associated with the ``"masses"`` or ``"masses/<variant>"`` name (see
+:ref:`output-variants`), and must have the following metadata:
 
 .. list-table:: Metadata for masses
   :widths: 2 3 7
@@ -36,9 +36,9 @@ inputs, and must adhere to the following metadata schema:
     - the masses must not have any components
 
   * - properties
-    - ``"masses"``
-    - masses must have a single property dimension named
-      ``"masses"``, with a single entry set to ``0``.
+    - ``"mass``
+    - masses must have a single property dimension named ``"mass"``, with a
+      single entry set to ``0``.
 
 The following simulation engine can provide ``"masses"`` as inputs to the models.
 
@@ -51,4 +51,3 @@ The following simulation engine can provide ``"masses"`` as inputs to the models
     :link-type: ref
 
     |ase-logo|
-

docs/src/outputs/momenta.rst

@@ -8,8 +8,8 @@ Predictions of momenta can be used, for example, to predict a future step in mol
 dynamics (see, e.g., https://arxiv.org/pdf/2505.19350).
 
 In metatomic models, they are associated with the ``"momenta"`` or
-``"momenta/<variant>"`` key (see :ref:`output-variants`) key in the model
-outputs, and must adhere to the following metadata schema:
+``"momenta/<variant>"`` name (see :ref:`output-variants`), and must have the
+following metadata:
 
 .. list-table:: Metadata for momenta
   :widths: 2 3 7
@@ -44,9 +44,9 @@ outputs, and must adhere to the following metadata schema:
       components is x, y, z.
 
   * - properties
-    - ``"momenta"``
+    - ``"momentum"``
     - momenta must have a single property dimension named
-      ``"momenta"``, with a single entry set to ``0``.
+      ``"momentum"``, with a single entry set to ``0``.
 
 The following simulation engine can provide ``"momenta"`` as inputs to the models.
 
@@ -59,4 +59,3 @@ The following simulation engine can provide ``"momenta"`` as inputs to the model
     :link-type: ref
 
     |ase-logo|
-

docs/src/outputs/non_conservative.rst

@@ -8,9 +8,9 @@ gradient of a potential energy function. These are generally faster to compute
 than forces derived from the potential energy by backpropagation. However, these
 predictions must be used with care, see https://arxiv.org/abs/2412.11569.
 
-In metatomic models, they are associated with the ``"non_conservative_forces"`` or
-``"non_conservative_forces/<variant>"`` key (see :ref:`output-variants`) in the model
-outputs, and must adhere to the following metadata schema:
+In metatomic models, they are associated with the ``"non_conservative_forces"``
+or ``"non_conservative_forces/<variant>"`` name (see :ref:`output-variants`),
+and must have the following metadata:
 
 .. list-table:: Metadata for non-conservative forces
   :widths: 2 3 7
@@ -45,9 +45,9 @@ outputs, and must adhere to the following metadata schema:
       components is x, y, z.
 
   * - properties
-    - ``"non_conservative_forces"``
+    - ``"non_conservative_force"``
     - non-conservative forces must have a single property dimension named
-      ``"non_conservative_forces"``, with a single entry set to ``0``.
+      ``"non_conservative_force"``, with a single entry set to ``0``.
 
 The following simulation engines can use the ``"non_conservative_forces"``
 output, using a ``non_conservative`` flag:
@@ -95,9 +95,9 @@ not calculated using derivatives of the potential energy. As with forces, they
 are typically faster to compute but need to be used with care, see
 https://arxiv.org/abs/2412.11569.
 
-In metatomic models, they are associated with the ``"non_conservative_stress"`` or
-``"non_conservative_stress/<variant>"`` key (see :ref:`output-variants`) in the model
-outputs, and must adhere to the following metadata schema:
+In metatomic models, they are associated with the ``"non_conservative_stress"``
+or ``"non_conservative_stress/<variant>"`` name (see :ref:`output-variants`),
+and must have the following metadata:
 
 .. list-table:: Metadata for non-conservative stress output
   :widths: 2 3 7

docs/src/outputs/positions.rst

@@ -8,8 +8,8 @@ They can be used to predict the next configuration in molecular dynamics
 (see, e.g., https://arxiv.org/pdf/2505.19350).
 
 In metatomic models, they are associated with the ``"positions"`` or
-``"positions/<variant>"`` key (see :ref:`output-variants`) key in the model
-outputs, and must adhere to the following metadata schema:
+``"positions/<variant>"`` name (see :ref:`output-variants`), and must have the
+following metadata:
 
 .. list-table:: Metadata for positions
   :widths: 2 3 7
@@ -44,8 +44,8 @@ outputs, and must adhere to the following metadata schema:
       components is x, y, z.
 
   * - properties
-    - ``"positions"``
+    - ``"position"``
     - positions must have a single property dimension named
-      ``"positions"``, with a single entry set to ``0``.
+      ``"position"``, with a single entry set to ``0``.
 
 At the moment, positions are not integrated into any simulation engines.

docs/src/outputs/velocities.rst

@@ -3,8 +3,9 @@
 Velocities
 ^^^^^^^^^^
 
-Velocities are associated with the ``"velocities"`` key in the model
-inputs, and must adhere to the following metadata schema:
+Velocities are associated with the ``"velocities"`` or
+``"velocities/<variant>"`` name (see :ref:`output-variants`), and must have the
+following metadata:
 
 .. list-table:: Metadata for velocities
   :widths: 2 3 7
@@ -39,9 +40,9 @@ inputs, and must adhere to the following metadata schema:
       components is x, y, z.
 
   * - properties
-    - ``"velocities"``
+    - ``"velocity"``
     - velocities must have a single property dimension named
-      ``"velocities"``, with a single entry set to ``0``.
+      ``"velocity"``, with a single entry set to ``0``.
 
 The following simulation engine can provide ``"velocities"`` as inputs to the models.
 

metatomic-torch/CHANGELOG.md

@@ -17,6 +17,12 @@ a changelog](https://keepachangelog.com/en/1.1.0/) format. This project follows
 ### Removed
 -->
 
+### Changed
+
+- Properties in standard output/inputs should now consistently use singular
+  (i.e. `position` instead of `positions`). The plural names are deprecated and
+  will be remove in the future.
+
 ## [Version 0.1.7](https://github.com/metatensor/metatomic/releases/tag/metatomic-torch-v0.1.7) - 2025-11-26
 
 ### Fixed

metatomic-torch/src/outputs.cpp

@@ -326,9 +326,26 @@ static void check_non_conservative_forces(
             torch::tensor({{0}, {1}, {2}}, tensor_options)
         )
     };
-
     validate_components("non_conservative_forces", forces_block->components(), expected_components);
 
+    Labels expected_properties;
+    if (forces_block->properties()->names()[0] == "non_conservative_forces") {
+        TORCH_WARN_ONCE(
+            "The 'non_conservative_forces' output uses a deprecated property name "
+            "'non_conservative_forces'. Please use 'non_conservative_force' (singular) instead."
+        )
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "non_conservative_forces",
+            torch::tensor({{0}}, tensor_options)
+        );
+    } else {
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "non_conservative_force",
+            torch::tensor({{0}}, tensor_options)
+        );
+    }
+    validate_properties("non_conservative_forces", forces_block, expected_properties);
+
     // Should not have any gradients
     validate_no_gradients("non_conservative_forces", forces_block);
 }
@@ -355,6 +372,12 @@ static void check_non_conservative_stress(
 
     validate_components("non_conservative_stress", stress_block->components(), expected_components);
 
+    auto expected_properties = torch::make_intrusive<LabelsHolder>(
+        "non_conservative_stress",
+        torch::tensor({{0}}, tensor_options)
+    );
+    validate_properties("non_conservative_stress", stress_block, expected_properties);
+
     // Should not have any gradients
     validate_no_gradients("non_conservative_stress", stress_block);
 }
@@ -382,10 +405,22 @@ static void check_positions(
 
     validate_components("positions", positions_block->components(), expected_components);
 
-    auto expected_properties = torch::make_intrusive<LabelsHolder>(
-        "positions",
-        torch::tensor({{0}}, tensor_options)
-    );
+    Labels expected_properties;
+    if (positions_block->properties()->names()[0] == "positions") {
+        TORCH_WARN_ONCE(
+            "The 'positions' output uses a deprecated property name 'positions'. "
+            "Please use 'position' (singular) instead."
+        )
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "positions",
+            torch::tensor({{0}}, tensor_options)
+        );
+    } else {
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "position",
+            torch::tensor({{0}}, tensor_options)
+        );
+    }
     validate_properties("positions", positions_block, expected_properties);
 
     // Should not have any gradients
@@ -404,7 +439,6 @@ static void check_momenta(
     // Check samples values from systems
     validate_atomic_samples("momenta", value, systems, request, torch::nullopt);
 
-
     auto tensor_options = torch::TensorOptions().device(value->device());
     auto momenta_block = TensorMapHolder::block_by_id(value, 0);
     std::vector<Labels> expected_component {
@@ -415,10 +449,22 @@ static void check_momenta(
     };
     validate_components("momenta", momenta_block->components(), expected_component);
 
-    auto expected_properties = torch::make_intrusive<LabelsHolder>(
-        "momenta",
-        torch::tensor({{0}}, tensor_options)
-    );
+    Labels expected_properties;
+    if (momenta_block->properties()->names()[0] == "momenta") {
+        TORCH_WARN_ONCE(
+            "The 'momenta' output uses a deprecated property name 'momenta'. "
+            "Please use 'momentum' (singular) instead."
+        )
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "momenta",
+            torch::tensor({{0}}, tensor_options)
+        );
+    } else {
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "momentum",
+            torch::tensor({{0}}, tensor_options)
+        );
+    }
     validate_properties("momenta", momenta_block, expected_properties);
 
     // Should not have any gradients
@@ -443,10 +489,22 @@ static void check_masses(
     // Ensure that the block has no components
     validate_components("masses", masses_block->components(), {});
 
-    auto expected_properties = torch::make_intrusive<LabelsHolder>(
-        "masses",
-        torch::tensor({{0}}, tensor_options)
-    );
+    Labels expected_properties;
+    if (masses_block->properties()->names()[0] == "masses") {
+        TORCH_WARN_ONCE(
+            "The 'masses' output uses a deprecated property name 'masses'. "
+            "Please use 'mass' (singular) instead."
+        )
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "masses",
+            torch::tensor({{0}}, tensor_options)
+        );
+    } else {
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "mass",
+            torch::tensor({{0}}, tensor_options)
+        );
+    }
     validate_properties("masses", masses_block, expected_properties);
 
     // Should not have any gradients
@@ -475,10 +533,23 @@ static void check_velocities(
     };
     validate_components("velocities", velocities_block->components(), expected_component);
 
-    auto expected_properties = torch::make_intrusive<LabelsHolder>(
-        "velocities",
-        torch::tensor({{0}}, tensor_options)
-    );
+    Labels expected_properties;
+    if (velocities_block->properties()->names()[0] == "velocities") {
+        TORCH_WARN_ONCE(
+            "The 'velocities' output uses a deprecated property name 'velocities'. "
+            "Please use 'velocity' (singular) instead."
+        )
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "velocities",
+            torch::tensor({{0}}, tensor_options)
+        );
+    } else {
+        expected_properties = torch::make_intrusive<LabelsHolder>(
+            "velocity",
+            torch::tensor({{0}}, tensor_options)
+        );
+    }
+
     validate_properties("velocities", velocities_block, expected_properties);
 
     // Should not have any gradients
@@ -504,7 +575,7 @@ static void check_charges(
     validate_components("charges", charges_block->components(), {});
 
     auto expected_properties = torch::make_intrusive<LabelsHolder>(
-        "charges",
+        "charge",
         torch::tensor({{0}}, tensor_options)
     );
     validate_properties("charges", charges_block, expected_properties);

python/metatomic_torch/metatomic/torch/ase_calculator.py

@@ -998,6 +998,10 @@ def _get_ase_input(
     # for metatensor
     values = torch.tensor(values[..., None])
 
+    components = []
+    if values.shape[1] != 1:
+        components.append(Labels(["xyz"], torch.arange(values.shape[1]).reshape(-1, 1)))
+
     block = TensorBlock(
         values,
         samples=Labels(
@@ -1006,15 +1010,8 @@ def _get_ase_input(
                 [torch.full((values.shape[0],), 0), torch.arange(values.shape[0])]
             ).T,
         ),
-        components=[Labels(["xyz"], torch.arange(values.shape[1]).reshape(-1, 1))]
-        if values.shape[1] != 1
-        else [],
-        properties=Labels(
-            [
-                name if "::" not in name else name.split("::")[-1],
-            ],
-            torch.tensor([[0]]),
-        ),
+        components=components,
+        properties=Labels([infos["quantity"]], torch.tensor([[0]])),
     )
 
     tensor = TensorMap(Labels(["_"], torch.tensor([[0]])), [block])