diff --git a/onnxscript/function_libs/torch_lib/ops/core.py b/onnxscript/function_libs/torch_lib/ops/core.py
index 27d30e377f..22abeee6d2 100644
--- a/onnxscript/function_libs/torch_lib/ops/core.py
+++ b/onnxscript/function_libs/torch_lib/ops/core.py
@@ -8029,6 +8029,16 @@ def aten_pow_tensor_scalar(self: TReal, exponent: float) -> TReal:
 @torch_op("aten::pow.Scalar", trace_only=True)
 def aten_pow_scalar(self: float, exponent: TTensor) -> TTensor:
     """pow.Scalar(Scalar self, Tensor exponent) -> Tensor"""
+    if not isinstance(self, int) and not exponent.dtype.is_floating_point():
+        # A float scalar outranks an integral exponent, so torch promotes the result to
+        # the default float type instead of narrowing the scalar down to the exponent
+        return op.Pow(op.Cast(self, to=FLOAT.dtype), op.Cast(exponent, to=FLOAT.dtype))
+    if exponent.dtype == ir.DataType.BOOL:
+        # Pow has no boolean inputs, and an int scalar over a boolean exponent
+        # promotes to the default int type in torch
+        return op.Pow(op.Cast(self, to=INT64.dtype), op.Cast(exponent, to=INT64.dtype))
+    # The exponent is in the same or a higher type category than the scalar, so it
+    # decides the result type. e.g. 2.0 ** float16 tensor is float16
     return op.Pow(op.Cast(self, to=exponent.dtype), exponent)
 
 
diff --git a/tests/function_libs/torch_lib/e2e_ops_tests.py b/tests/function_libs/torch_lib/e2e_ops_tests.py
index 8a50b5d58d..bfbe33e434 100644
--- a/tests/function_libs/torch_lib/e2e_ops_tests.py
+++ b/tests/function_libs/torch_lib/e2e_ops_tests.py
@@ -84,6 +84,65 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
         )
         _testing.assert_onnx_program(onnx_program)
 
+    def test_pow_scalar_float_int(self):
+        class PowModel(torch.nn.Module):
+            def forward(self, x: torch.Tensor) -> torch.Tensor:
+                return 2.0**x
+
+        onnx_program = torch.onnx.export(
+            PowModel(),
+            (torch.tensor([1, 2, 3], dtype=torch.int64),),
+            dynamo=True,
+            optimize=False,
+        )
+        _testing.assert_onnx_program(onnx_program)
+
+    def test_pow_scalar_float_bool(self):
+        class PowModel(torch.nn.Module):
+            def forward(self, x: torch.Tensor) -> torch.Tensor:
+                return 2.0**x
+
+        onnx_program = torch.onnx.export(
+            PowModel(), (torch.tensor([True, False]),), dynamo=True, optimize=False
+        )
+        _testing.assert_onnx_program(onnx_program)
+
+    def test_pow_scalar_float_float16(self):
+        class PowModel(torch.nn.Module):
+            def forward(self, x: torch.Tensor) -> torch.Tensor:
+                return 2.0**x
+
+        onnx_program = torch.onnx.export(
+            PowModel(),
+            (torch.tensor([1.0, 2.0], dtype=torch.float16),),
+            dynamo=True,
+            optimize=False,
+        )
+        _testing.assert_onnx_program(onnx_program)
+
+    def test_pow_scalar_int_int(self):
+        class PowModel(torch.nn.Module):
+            def forward(self, x: torch.Tensor) -> torch.Tensor:
+                return 2**x
+
+        onnx_program = torch.onnx.export(
+            PowModel(),
+            (torch.tensor([1, 2, 3], dtype=torch.int64),),
+            dynamo=True,
+            optimize=False,
+        )
+        _testing.assert_onnx_program(onnx_program)
+
+    def test_pow_scalar_int_bool(self):
+        class PowModel(torch.nn.Module):
+            def forward(self, x: torch.Tensor) -> torch.Tensor:
+                return 2**x
+
+        onnx_program = torch.onnx.export(
+            PowModel(), (torch.tensor([True, False]),), dynamo=True, optimize=False
+        )
+        _testing.assert_onnx_program(onnx_program)
+
     def test_mul_tensor_scalar_float(self):
         class Model(torch.nn.Module):
             def forward(self, x: torch.Tensor) -> torch.Tensor: