Numba Split: Validate sizes

pytensor/link/numba/dispatch/tensor_basic.py

@@ -124,10 +124,19 @@ def numba_funcify_Join(op, **kwargs):
 @register_funcify_default_op_cache_key(Split)
 def numba_funcify_Split(op, **kwargs):
     @numba_basic.numba_njit
-    def split(tensor, axis, indices):
-        return np.split(tensor, np.cumsum(indices)[:-1], axis=axis.item())
+    def split(x, axis, sizes):
+        if (sizes < 0).any():
+            raise ValueError("Split sizes cannot be negative")
+        axis = axis.item()
+        split_indices = np.cumsum(sizes)
+        if split_indices[-1] != x.shape[axis]:
+            raise ValueError(
+                f"Split sizes sum to {split_indices[-1]}; expected {x.shape[axis]}"
+            )
+        return np.split(x, split_indices[:-1], axis=axis)
 
-    return split
+    cache_version = 1
+    return split, cache_version
 
 
 @register_funcify_default_op_cache_key(ExtractDiag)

tests/link/numba/test_tensor_basic.py

@@ -10,6 +10,7 @@ from pytensor.scalar import Add
 from tests.link.numba.test_basic import (
     compare_numba_and_py,
     compare_shape_dtype,
+    numba_mode,
 )
 from tests.tensor.test_basic import check_alloc_runtime_broadcast
 
@@ -245,6 +246,18 @@ def test_Split_view():
     )
 
 
+def test_split_errors():
+    x = pt.dvector("x", shape=(5,))
+    splits = pt.tensor(shape=(3,), dtype="int64")
+    outs = pt.split(x, splits)
+    fn = function([x, splits], outs, mode=numba_mode)
+    test_x = np.zeros((5,))
+    with pytest.raises(ValueError, match="Split sizes sum to 4; expected 5"):
+        fn(test_x, np.array([1, 2, 1], dtype="int64"))
+    with pytest.raises(ValueError, match="Split sizes cannot be negative"):
+        fn(test_x, np.array([2, 4, -1], dtype="int64"))
+
+
 @pytest.mark.parametrize(
     "val, offset",
     [