[PyTorch] Fix rsub type (#10090)

* [PyTorch] Fix rsub type

* fix

Author: comaniac

python/tvm/relay/frontend/pytorch.py

@@ -1972,10 +1972,7 @@ def stack(self, inputs, input_types):
             return self.tensor_array_stack(inputs, input_types)
 
     def rsub(self, inputs, input_types):
-        data0, data1 = self.pytorch_promote_types(inputs[:2], input_types[:2])
-
-        # TODO (t-vi): should this also be part of the type promotion?
-        alpha = _expr.const(float(inputs[2]))
+        data0, data1, alpha = self.pytorch_promote_types(inputs, input_types)
 
         # note: rsub means data0 and data1 swap places
         return get_relay_op("subtract")(data1, alpha * data0)

tests/python/frontend/pytorch/test_forward.py

@@ -2691,6 +2691,13 @@ def forward(self, *args):
     verify_model(Rsub2().float().eval(), input_data=[d1, d2])
     verify_model(Rsub2().float().eval(), input_data=[d1, d3])
 
+    d1 = torch.rand([1, 3]).half()
+    d2 = torch.rand([1, 3]).half()
+    verify_model(Rsub1().half().eval(), input_data=[d1, d2])
+    verify_model(Rsub1().half().eval(), input_data=[d1, d3])
+    verify_model(Rsub2().half().eval(), input_data=[d1, d2])
+    verify_model(Rsub2().half().eval(), input_data=[d1, d3])
+
 
 @tvm.testing.uses_gpu
 def test_forward_embedding():