fix(pruning): validate multidimensional tensor in channel-structured pruning (#107)

changelog.d/107.fixed

@@ -0,0 +1 @@
+Validate that channel-structured magnitude pruning is applied to tensors with at least 2 dimensions, raising a descriptive ValueError for 1D tensors

src/coreai_opt/pruning/spec/prune.py

@@ -139,14 +139,15 @@ def compute_mask(
         Returns:
             torch.Tensor: Binary mask (1 = keep, 0 = prune).
         """
+        # TODO: Replace this with generic abstractions
+        if isinstance(pruning_scheme, ChannelStructured):
+            return _MagnitudePruneImpl._compute_channel_mask(weight, sparsity, pruning_scheme.axis)
+
         if sparsity == 0.0:
             return torch.ones_like(weight)
         if sparsity >= 1.0:
             return torch.zeros_like(weight)
 
-        # TODO: Replace this with generic abstractions
-        if isinstance(pruning_scheme, ChannelStructured):
-            return _MagnitudePruneImpl._compute_channel_mask(weight, sparsity, pruning_scheme.axis)
         return _MagnitudePruneImpl._compute_unstructured_mask(weight, sparsity)
 
     @staticmethod
@@ -171,6 +172,13 @@ def _compute_channel_mask(
         Channel importance is measured by L1 norm. The least-important
         channels are pruned entirely.
         """
+        if weight.ndim < 2:
+            raise ValueError(
+                f"Channel-structured pruning requires a tensor with at least 2 dimensions, "
+                f"got shape {tuple(weight.shape)} with {weight.ndim} dims. "
+                f"For 1D tensors, use Unstructured pruning instead."
+            )
+
         if not (-weight.ndim <= axis < weight.ndim):
             raise ValueError(
                 f"Invalid axis. Should be in range [{-weight.ndim}, {weight.ndim}), but got {axis}"

tests/pruning/test_magnitude_pruner.py

@@ -23,7 +23,12 @@
     OpMagnitudePrunerConfig,
     PolynomialDecaySchedule,
 )
-from coreai_opt.pruning.spec import ChannelStructured, PruneImplBase, Unstructured
+from coreai_opt.pruning.spec import (
+    ChannelStructured,
+    PruneImplBase,
+    Unstructured,
+    _MagnitudePruneImpl,
+)
 
 
 @pytest.fixture
@@ -462,6 +467,18 @@ def test_channel_structured_axis_out_of_range(self, axis: int) -> None:
         with pytest.raises(ValueError, match="Invalid axis"):
             pruner.prepare((torch.randn(1, 4),))
 
+    @pytest.mark.parametrize("target_sparsity", [0.0, 0.5])
+    @pytest.mark.parametrize("axis", [0, -1], ids=["axis-0", "axis-neg-1"])
+    def test_channel_structured_1d_tensor_raises(self, axis: int, target_sparsity: float) -> None:
+        """Channel-structured pruning requires >= 2 dimensions; 1D tensors raise ValueError."""
+        weight = torch.tensor([1.0, 5.0, 2.0, 8.0])
+        scheme = ChannelStructured(axis=axis)
+        with pytest.raises(
+            ValueError,
+            match=r"Channel-structured pruning requires a tensor with at least 2 dimensions",
+        ):
+            _MagnitudePruneImpl.compute_mask(weight, target_sparsity, scheme)
+
     def test_linear_unstructured_conv2d_channel_structured(self) -> None:
         """Apply unstructured to Linear and channel-structured to Conv2d in same model."""