Add reduce_max_grad composite rule #51653
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
|
pr name typo |
wangxn12138
changed the title
done |
| def test_check_grad(self): | ||
| # only composite op support gradient check of reduce_max | ||
| self.check_grad(['X'], 'Out', check_eager=True, only_check_prim=True) | ||
|
|
||
| def test_raise_error(self): |
There was a problem hiding this comment.
if original test pass fp16, please describe it in pr describes
There was a problem hiding this comment.
if original test pass fp16, please describe it in pr describes
done
| @@ -271,6 +277,10 @@ def setUp(self): | |||
| def test_check_output(self): | |||
| self.check_output(check_eager=True) | |||
|
|
|||
| def test_check_grad(self): | |||
There was a problem hiding this comment.
why original op test don't support grad check
There was a problem hiding this comment.
@Charles-hit if original op has no grad test how to compare with composite_grad
There was a problem hiding this comment.
why original op test don't support grad check
装饰器注明"reduce_max is discontinuous non-derivable function, its gradient check is not supported by unittest framework."
| @@ -261,6 +266,7 @@ class TestMaxOp_ZeroDim(OpTest): | |||
|
|
|||
| def setUp(self): | |||
| self.op_type = "reduce_max" | |||
| self.prim_op_type = "prim" | |||
| self.python_api = paddle.max | |||
| self.inputs = {'X': np.random.random([]).astype("float64")} | |||
| self.attrs = {'dim': []} | |||
There was a problem hiding this comment.
it's better add new op test to check keep_dim = True and FP32
There was a problem hiding this comment.
it's better add new op test to check keep_dim = True and FP32
done
| @@ -233,6 +233,7 @@ class TestMaxOp(OpTest): | |||
|
|
|||
| def setUp(self): | |||
| self.op_type = "reduce_max" | |||
| self.prim_op_type = "prim" | |||
| self.python_api = paddle.max | |||
| self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")} | |||
| self.attrs = {'dim': [-1]} | |||
There was a problem hiding this comment.
it's better to check keep_dim = True
There was a problem hiding this comment.
it's better to check keep_dim = True
done
paddle/fluid/prim/api/api.yaml
Outdated
| @@ -37,6 +37,8 @@ | |||
| - pad | |||
| - cumsum | |||
| - put_along_axis | |||
| - equal | |||
| - where | |||
There was a problem hiding this comment.
two where， delete one
There was a problem hiding this comment.
two where， delete one
done
PR types
New features
PR changes
Others
Describe
add reduce_max_grad composite rule.
becuse original op test pass fp16, composite op test pass fp16 too.