Remove paddle.where in cross_tropy_loss

[HydrogenSulfate]

PR types

Function optimization

PR changes

APIs

Describe

1. 使用已修复BUG的paddle.min和paddle.max代替paddle.where来检查labels合法性，减少GPU和CPU之间不必要的同步

[paddle-bot-old]

✅ This PR's description meets the template requirements!
Please wait for other CI results.

[paddle-bot-old]

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

[pangyoki]

可以描述下删掉的这些check的功能和删掉之后有啥影响吗？为什么之前需要加上这个checking，现在直接删掉的时候，出现这些case是有其它代码会检查吗？

[HydrogenSulfate]

可以描述下删掉的这些check的功能和删掉之后有啥影响吗？为什么之前需要加上这个checking，现在直接删掉的时候，出现这些case是有其它代码会检查吗？

去掉这个检查后，如果labels范围超出边界，则会在gather_nd这里报错

def test_LabelValue_ExceedMax():
                input_data = paddle.rand(shape=[20, 100])
                label_data = paddle.randint(
                    0, 100, shape=[20, 1], dtype="int64")  # hard label
                label_data[0] = 100
                weight_data = paddle.rand([100])  # provide weight
                paddle.nn.functional.cross_entropy(
                    input=input_data,
                    label=label_data,
                    weight=weight_data,
                    ignore_index=-100)


Error: /paddle/paddle/fluid/operators/gather.cu.h:62 Assertion `index_value >= 0 && index_value < input_dims[j]` failed. The index is out of bounds, please check whether the dimensions of index and input meet the requirements. It should be less than [100] and greater than or equal to 0, but received [0]

def test_LabelValue_ExceedMin():
                input_data = paddle.rand(shape=[20, 100])
                label_data = paddle.randint(
                    0, 100, shape=[20, 1], dtype="int64")  # hard label
                label_data[0] = -1
                weight_data = paddle.rand([100])  # provide weight
                paddle.nn.functional.cross_entropy(
                    input=input_data,
                    label=label_data,
                    weight=weight_data,
                    ignore_index=-100)

Error: /paddle/paddle/fluid/operators/gather.cu.h:62 Assertion `index_value >= 0 && index_value < input_dims[j]` failed. The index is out of bounds, please check whether the dimensions of index and input meet the requirements. It should be less than [100] and greater than or equal to 0, but received [0]

[pangyoki]

可以描述下删掉的这些check的功能和删掉之后有啥影响吗？为什么之前需要加上这个checking，现在直接删掉的时候，出现这些case是有其它代码会检查吗？

去掉这个检查后，如果labels范围超出边界，则会在gather_nd这里报错

def test_LabelValue_ExceedMax():
                input_data = paddle.rand(shape=[20, 100])
                label_data = paddle.randint(
                    0, 100, shape=[20, 1], dtype="int64")  # hard label
                label_data[0] = 100
                weight_data = paddle.rand([100])  # provide weight
                paddle.nn.functional.cross_entropy(
                    input=input_data,
                    label=label_data,
                    weight=weight_data,
                    ignore_index=-100)


Error: /paddle/paddle/fluid/operators/gather.cu.h:62 Assertion `index_value >= 0 && index_value < input_dims[j]` failed. The index is out of bounds, please check whether the dimensions of index and input meet the requirements. It should be less than [100] and greater than or equal to 0, but received [0]

def test_LabelValue_ExceedMin():
                input_data = paddle.rand(shape=[20, 100])
                label_data = paddle.randint(
                    0, 100, shape=[20, 1], dtype="int64")  # hard label
                label_data[0] = -1
                weight_data = paddle.rand([100])  # provide weight
                paddle.nn.functional.cross_entropy(
                    input=input_data,
                    label=label_data,
                    weight=weight_data,
                    ignore_index=-100)

Error: /paddle/paddle/fluid/operators/gather.cu.h:62 Assertion `index_value >= 0 && index_value < input_dims[j]` failed. The index is out of bounds, please check whether the dimensions of index and input meet the requirements. It should be less than [100] and greater than or equal to 0, but received [0]

这样是不是就不容易判断出是cross_entropy的报错了啊，有python栈可以看出是cross_entropy的问题吗？
静态图也是相同的报错吗？是应该把动态图位置的报错删掉，还是说应该为静态图加上报错呢？

[paddle-bot-old]

Sorry to inform you that 056394c's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.

[pangyoki]

之前讨论过在weight is not None的条件下，才需要判断label的合法性。需要把检查移动到weight is not None中吗？

[HydrogenSulfate]

由于softmax_with_cross_entropy中没有对labels进行充分的合法性检查，所以这里在hard label条件下就判断label的合法性，无论是否用了weight

[pangyoki]

LGTM

[chajchaj]

LGTM