[Bug fixes] Add default arg to enhance varbase ClearGradient func

[veyron95]

PR types

Bug fixes

PR changes

APIs

Describe

1. Add arg to enhance varbase ClearGradient func,

use case: 1

 linear.weight.clear_gradient() #  same as the original version.

use case: 2

 linear.weight.clear_gradient(False)   # False means to real clear gradient
 # default args: set_to_zero = True

2. Expose two func to set and get the " is_empty_ " attribute of grad_var_

• _is_gradient_set_empty()
• _gradient_set_empty(is_empty = False) or _gradient_set_empty(is_empty = True) , default=True

@dygraph_guard
    def test_tensor_method_clear_gradient_case2(self):
        input = paddle.uniform(self.input_shape)
        linear = paddle.nn.Linear(2, 3)
        out = linear(input)
        out.backward()
        # default arg set_to_zero is true
        # so, False means real clear gradient
        linear.weight.clear_gradient(False)

        # before ._gradient_set_empty(False), 
        # the return of ._is_gradient_set_empty() should be True
        self.assertTrue(linear.weight._is_gradient_set_empty())

        # reset, because ClearGradient will call SetIsEmpty(True), but this is not our expectation.
        linear.weight._gradient_set_empty(False)
        # after ._gradient_set_empty(False), 
        # the return of ._is_gradient_set_empty() should be False
        self.assertFalse(linear.weight._is_gradient_set_empty())

        # actual result
        gradient_actual = linear.weight.grad
        # expected result
        self.assertTrue(np.empty(gradient_actual))

[paddle-bot-old]

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[Baibaifan]

LGTM for collective

[paddle-bot-old]

Sorry to inform you that 9c458ff's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.

[JiabinYang]

some comments

[TCChenlong]

LGTM

[JiabinYang]

LGTM