remove copy_

python/tvm/relay/frontend/pytorch.py

@@ -1756,15 +1756,6 @@ def _impl(inputs, input_types):
     return _impl
 
 
-def _copy_():
-    def _impl(inputs, input_types):
-        # use add to help handle broadcasting
-        rel = _op.zeros_like(inputs[0])
-        return _op.add(rel, inputs[1])
-
-    return _impl
-
-
 def _none():
     def _impl(inputs, input_types):
         return None
@@ -2641,7 +2632,6 @@ def _get_convert_map(prelude, default_dtype):
         "aten::isnan": _unary("isnan"),
         "aten::clamp": _clamp(),
         "aten::clamp_": _clamp(),
-        "aten::copy_": _copy_(),
         "aten::detach": _identity(),
         "aten::upsample_bilinear2d": _upsample("bilinear", prelude),
         "aten::upsample_nearest2d": _upsample("nearest_neighbor", prelude),

tests/python/frontend/pytorch/test_forward.py

@@ -2416,34 +2416,6 @@ def forward(self, *args):
         verify_model(ClampInPlace(min, max).float().eval(), input_data=input_data)
 
 
-@tvm.testing.uses_gpu
-def test_forward_copy_():
-    torch.set_grad_enabled(False)
-
-    class Copy(Module):
-        def __init__(self):
-            super(Copy, self).__init__()
-
-        def forward(self, *args):
-            return torch.Tensor.copy_(args[0], args[1])
-
-    class CopyInPlace(Module):
-        def __init__(self):
-            super(CopyInPlace, self).__init__()
-
-        def forward(self, *args):
-            a = args[0]
-            b = args[1]
-            c = torch.Tensor.copy_(a, b)
-            return a
-
-    src_tensor = torch.rand(5)
-    tgt_tensor = torch.rand((2, 3, 5))
-    for copy in [Copy, CopyInPlace]:
-        verify_model(copy().float().eval(), input_data=[tgt_tensor, src_tensor])
-        verify_model(copy().float().eval(), input_data=[tgt_tensor, src_tensor + tgt_tensor])
-
-
 @tvm.testing.uses_gpu
 def test_forward_ones():
     torch.set_grad_enabled(False)
@@ -2940,7 +2912,7 @@ def forward(self, *args):
     t2 = torch.rand([1, 3]).float()
     verify_model(Addcmul2().float().eval(), input_data=[input_data, t1, t2])
 
-
+@tvm.testing.uses_gpu
 def test_forward_true_divide():
     torch.set_grad_enabled(False)
 
@@ -2950,10 +2922,12 @@ def forward(self, *args):
 
     dividend = torch.rand([5, 3]).float()
     # divisor could be either tensor or scalar
-    divisor_tensor = torch.rand([5, 3]).float()
-    divisor_scalar = divisor = torch.tensor(1.0, dtype=torch.float32)
-    verify_model(TrueDivide().float().eval(), input_data=[dividend, divisor_tensor])
-    verify_model(TrueDivide().float().eval(), input_data=[dividend, divisor_scalar])
+    divisor_tensor = torch.rand([5, 3]).float() + 0.5
+    divisor_scalar = torch.tensor(1.0, dtype=torch.float32)
+    verify_model(TrueDivide().float().eval(),
+                 input_data=[dividend, divisor_tensor], atol=1e-4, rtol=1e-4)
+    verify_model(TrueDivide().float().eval(),
+                 input_data=[dividend, divisor_scalar], atol=1e-4, rtol=1e-4)
 
 
 @tvm.testing.uses_gpu
@@ -3386,7 +3360,6 @@ def test_forward_pretrained_bert_base_uncased():
     test_forward_unary()
     test_forward_clamp()
     test_forward_clamp_()
-    test_forward_copy_()
     test_forward_logical_not()
     test_forward_bitwise_not()
     test_forward_bitwise_xor()