Numba DimShuffle: validate squeeze

pytensor/link/numba/dispatch/elemwise.py

@@ -397,10 +397,11 @@ def numba_funcify_CAReduce(op, node, **kwargs):
 
 
 @register_funcify_default_op_cache_key(DimShuffle)
-def numba_funcify_DimShuffle(op, node, **kwargs):
+def numba_funcify_DimShuffle(op: DimShuffle, node, **kwargs):
     # We use `as_strided` to achieve the DimShuffle behavior of transposing and expanding/squezing dimensions in one call
     # Numba doesn't currently support multiple expand/squeeze, and reshape is limited to contiguous arrays.
     new_order = tuple(op._new_order)
+    drop = tuple(op.drop)
     shape_template = (1,) * node.outputs[0].ndim
     strides_template = (0,) * node.outputs[0].ndim
 
@@ -409,6 +410,11 @@ def numba_funcify_DimShuffle(op, node, **kwargs):
 
         @numba_basic.numba_njit
         def squeeze_to_0d(x):
+            if not x.size == 1:
+                raise ValueError(
+                    "DimShuffle: Attempting to squeeze axes with size not equal to one"
+                )
+            assert x.size == 1
             return as_strided(x, shape=(), strides=())
 
         return squeeze_to_0d
@@ -428,10 +434,17 @@ def numba_funcify_DimShuffle(op, node, **kwargs):
                     new_strides = numba_basic.tuple_setitem(
                         new_strides, i, old_strides[o]
                     )
+            if drop:
+                for dropped_dim in drop:
+                    if old_shape[dropped_dim] != 1:
+                        raise ValueError(
+                            "DimShuffle: Attempting to squeeze axes with size not equal to one"
+                        )
 
             return as_strided(x, shape=new_shape, strides=new_strides)
 
-    return dimshuffle
+    cache_version = 1
+    return dimshuffle, cache_version
 
 
 @register_funcify_default_op_cache_key(Softmax)

tests/link/numba/test_elemwise.py

@@ -19,6 +19,7 @@ from pytensor.tensor.math import All, Any, Max, Min, Prod, ProdWithoutZeros, Sum
 from pytensor.tensor.special import LogSoftmax, Softmax, SoftmaxGrad
 from tests.link.numba.test_basic import (
     compare_numba_and_py,
+    numba_mode,
     scalar_my_multi_out,
 )
 from tests.tensor.test_elemwise import (
@@ -217,6 +218,17 @@ def test_Dimshuffle_non_contiguous():
     assert func(np.zeros(3), np.array([1])).ndim == 0
 
 
+def test_Dimshuffle_squeeze_errors():
+    x = pt.tensor3("x", shape=(4, None, 5))
+    out = pt.squeeze(x, axis=1)
+    assert out.type.shape == (4, 5)
+    fn = function([x], out, mode=numba_mode)
+    with pytest.raises(
+        ValueError, match="Attempting to squeeze axes with size not equal to one"
+    ):
+        fn(np.zeros((4, 2, 5)))
+
+
 @pytest.mark.parametrize(
     "careduce_fn, axis, v",
     [