diff --git a/_doc/sphinxdoc/source/api/tools.rst b/_doc/sphinxdoc/source/api/tools.rst
index de4bc88b7..ac51a43a6 100644
--- a/_doc/sphinxdoc/source/api/tools.rst
+++ b/_doc/sphinxdoc/source/api/tools.rst
@@ -126,7 +126,7 @@ the possibility later to only show a part of a graph.
 
 **benchmark**
 
-.. autosignature:: mlprodict.plotting.validate_graph.plot_validate_benchmark
+.. autosignature:: mlprodict.plotting.plot_validate_benchmark
 
 Others
 ======
diff --git a/_unittests/ut_onnxrt/test_onnxrt_python_runtime_.py b/_unittests/ut_onnxrt/test_onnxrt_python_runtime_.py
index 06bbbf60e..0f0936619 100644
--- a/_unittests/ut_onnxrt/test_onnxrt_python_runtime_.py
+++ b/_unittests/ut_onnxrt/test_onnxrt_python_runtime_.py
@@ -31,7 +31,7 @@
     OnnxBatchNormalization,
     OnnxAcos, OnnxAcosh, OnnxAsin, OnnxAsinh, OnnxAtan, OnnxAtanh,
     OnnxAveragePool,
-    OnnxCast, OnnxCeil, OnnxClip,
+    OnnxCast, OnnxCastLike, OnnxCeil, OnnxClip,
     OnnxCompress,
     OnnxConcat, OnnxConv, OnnxConvTranspose,
     OnnxConstant, OnnxConstant_9, OnnxConstant_11,
@@ -1083,6 +1083,25 @@ def test_onnxt_runtime_cast_in(self):
 
         python_tested.append(OnnxCast)
 
+    @wraplog()
+    def test_onnxt_runtime_cast_like(self):
+        x = numpy.array([1.5, 2.1, 3.1, 4.1]).astype(
+            numpy.float32)  # pylint: disable=E1101
+        y = numpy.array([1.]).astype(numpy.int64)  # pylint: disable=E1101
+
+        for opset in range(15, get_opset_number_from_onnx() + 1):
+            with self.subTest(opset=opset):
+                onx = OnnxCastLike('X', 'Y', output_names=['Z'],
+                                   op_version=opset)
+                model_def = onx.to_onnx(
+                    {'X': x, 'Y': y},
+                    outputs=[('Z', Int64TensorType())],
+                    target_opset=opset)
+                got = OnnxInference(model_def).run({'X': x, 'Y': y})
+                self.assertEqual(x.astype(numpy.int64), got['Z'])
+
+        python_tested.append(OnnxCastLike)
+
     @wraplog()
     def test_onnxt_runtime_ceil(self):
         self.common_test_onnxt_runtime_unary(OnnxCeil, numpy.ceil)
diff --git a/mlprodict/onnxrt/ops_cpu/_op_list.py b/mlprodict/onnxrt/ops_cpu/_op_list.py
index ff85fe9f2..f2b4d19f2 100644
--- a/mlprodict/onnxrt/ops_cpu/_op_list.py
+++ b/mlprodict/onnxrt/ops_cpu/_op_list.py
@@ -22,7 +22,7 @@
 from .op_batch_normalization import BatchNormalization, BatchNormalization_14
 from .op_binarizer import Binarizer
 from .op_broadcast_gradient_args import BroadcastGradientArgs
-from .op_cast import Cast
+from .op_cast import Cast, CastLike
 from .op_cdist import CDist
 from .op_ceil import Ceil
 from .op_celu import Celu
diff --git a/mlprodict/onnxrt/ops_cpu/op_cast.py b/mlprodict/onnxrt/ops_cpu/op_cast.py
index 2566302bc..0ce03ab9a 100644
--- a/mlprodict/onnxrt/ops_cpu/op_cast.py
+++ b/mlprodict/onnxrt/ops_cpu/op_cast.py
@@ -73,3 +73,29 @@ def _infer_types(self, x):  # pylint: disable=W0221
     def _infer_sizes(self, *args, **kwargs):
         res = self.run(*args, **kwargs)
         return (dict(temp=0), ) + res
+
+
+class CastLike(OpRun):
+
+    def __init__(self, onnx_node, desc=None, **options):
+        OpRun.__init__(self, onnx_node, desc=desc, **options)
+
+    def _run(self, x, y):  # pylint: disable=W0221
+        if self.inplaces.get(0, False):
+            return self._run_inplace(x, y)
+        return (x.astype(y.dtype), )
+
+    def _run_inplace(self, x, y):
+        if x.dtype == y._dtype:
+            return (x, )
+        return (x.astype(y.dtype), )
+
+    def _infer_shapes(self, x, y):  # pylint: disable=W0221
+        return (x.copy(dtype=y._dtype), )
+
+    def _infer_types(self, x, y):  # pylint: disable=W0221
+        return (y._dtype, )
+
+    def _infer_sizes(self, *args, **kwargs):
+        res = self.run(*args, **kwargs)
+        return (dict(temp=0), ) + res