【Paddle Tensor 第二期复数类型支持问题No.9、10】为`paddle.not_equal`、`paddle.equal`添加复数类型支持。 #69968

MrXnneHang · 2024-12-05T02:27:23Z

PR Category

User Experience

PR Types

Others

Description

为compare kernel添加复数支持：

not_equal
equal

偏特化EqualFunctor加入对<phi::dtype::complex<double>和<phi::dtype::complex<float>的计算逻辑。

error 1 (solved)

TypeError: (InvalidType) Type promotion only support calculations between floating-point numbers and between complex and real numbers. But got different data type x: int16, y: int8. (at /paddle/paddle/phi/common/type_promotion.h:229

这个问题是不支持int的自动类型提升。
https://github.com/HydrogenSulfate/array-api-tests/blob/paddle/array_api_tests/test_operators_and_elementwise_functions.py#1119
我把这一行换到手动类型转换之后就pass了。

error 2 (solved)

FAILED array_api_tests/test_operators_and_elementwise_functions.py::test_equal[__eq__(x1, x2)] - AssertionError: out=True, but should be (x1 == x2)=False [__eq__()]
  x1=0.2500000260770321, x2=0.2500000298023224
assert True == False
Falsifying example: test_equal(
    ctx=BinaryParamContext(<__eq__(x1, x2)>),
    data=data(...),
)
Draw 1 (x1): Tensor(shape=[], dtype=float64, place=Place(cpu), stop_gradient=True,
       0.25000003)
Draw 2 (x2): Tensor(shape=[], dtype=float32, place=Place(cpu), stop_gradient=True,
       0.25000003)

问题二似乎和提前的截断有关。测试里有个bug。
https://github.com/HydrogenSulfate/array-api-tests/blob/paddle/array_api_tests/test_operators_and_elementwise_functions.py#L1117

手动类型转换前left和right就把data截断了。后续对left和right做类型转换也太晚了。
应该利用promoted_dtype进行初始化而不是类型转换。

这样子似乎可以解决:

def test_equal(ctx, data):


    left = data.draw(ctx.left_strat, label=ctx.left_sym)
    right = data.draw(ctx.right_strat, label=ctx.right_sym)

    
    if not ctx.right_is_scalar:
        left_dtype = left.dtype
        right_dtype = right.dtype
        # We manually promote the dtypes as incorrect internal type promotion
        # could lead to false positives. For example
        #
        #     >>> xp.equal(
        #     ...     xp.asarray(1.0, dtype=xp.float32),
        #     ...     xp.asarray(1.00000001, dtype=xp.float64),
        #     ... )
        #
        # would erroneously be True if float64 downcasted to float32.
        promoted_dtype = dh.promotion_table[left_dtype, right_dtype]
        print(f"promoted_dtype:{promoted_dtype}")

        # 修改策略以确保生成的数据具有正确的 dtype
        left_strat = ctx.left_strat.filter(lambda x: x.dtype == promoted_dtype)
        right_strat = ctx.right_strat.filter(lambda x: x.dtype == promoted_dtype)

        left = data.draw(left_strat, label=ctx.left_sym)
        right = data.draw(right_strat, label=ctx.right_sym)

    out = ctx.func(left, right)

    binary_param_assert_dtype(ctx, left, right, out, xp.bool)
    binary_param_assert_shape(ctx, left, right, out)
    binary_param_assert_against_refimpl(
        ctx, left, right, out, "==", operator.eq, res_stype=bool
    )

涉及手动类型提升的测试似乎都应该这么改。

error 3 (solved) :

Paddle Tensor == Tensor,Tensor != Tensor 结果的shape和equal,not_equal不同。

手动调用Tensor.equal:

import torch
import numpy as np
import paddle
@pytest.mark.parametrize("ctx", make_binary_params("not_equal", dh.all_dtypes))
@given(data=st.data())
def test_not_equal(ctx, data):
    left = data.draw(ctx.left_strat, label=ctx.left_sym)
    right = data.draw(ctx.right_strat, label=ctx.right_sym)
    left_np = np.array(left)
    left_tensor = torch.tensor(left_np)
    right_np = np.array(right)
    right_tensor = torch.tensor(right_np)
    expected = (left_tensor != right_tensor).shape
    out = ctx.func(left, right)
    out_shape = paddle.to_tensor(left).not_equal(paddle.to_tensor(right)).shape
    binary_param_assert_dtype(ctx, left, right, out, xp.bool)
    binary_param_assert_shape(ctx, left, right, out_shape,expected=expected)
    if not ctx.right_is_scalar:
        # See test_equal note
        promoted_dtype = dh.promotion_table[left.dtype, right.dtype]
        left = xp.astype(left, promoted_dtype)
        right = xp.astype(right, promoted_dtype)
    binary_param_assert_against_refimpl(
        ctx, left, right, out, "!=", operator.ne, res_stype=bool
    )

paddle-bot · 2024-12-05T02:27:28Z

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

HydrogenSulfate

思路没问题，但是感觉复数这块的修改需要额外添加特化模板，应该不是简单注册算子就能解决的。
可以看到比较类型的Kernel确实是一起用宏生成的

以LessEqualFunctor为例，Kernel调用的Functor也是宏生成的，其逻辑在

也就是说核心比较代码是a op b，这对于实数是可以的，但是对于复数，应该是要比较其实部和虚部

另外看python和pytorch对于复数比较的支持情况来看，目前只支持了==，对于>, <, >=, <=都不支持，所以应该只需要为 == 的复数case写一个特化模板即可

paddle/phi/kernels/kps/compare_kernel.cu

…_equal` and `equal`

HydrogenSulfate

除了模板合并之外，其他感觉没问题了

HydrogenSulfate · 2024-12-05T09:29:00Z

paddle/phi/kernels/funcs/compare_functors.h

+template <typename OutT>
+struct EqualFunctor<phi::dtype::complex<float>, OutT> {
+  HOSTDEVICE OutT operator()(const phi::dtype::complex<float>& a,
+                             const phi::dtype::complex<float>& b) const {
+    if (isinf(a.real) || isinf(a.imag) || isinf(b.real) || isinf(b.imag)) {
+      return a == b;
+    }
+    if (isnan(a.real) || isnan(a.imag) || isnan(b.real) || isnan(b.imag)) {
+      return false;
+    }
+    float epsilon = 1e-8f;
+    return std::abs(a.real - b.real) < epsilon &&
+           std::abs(a.imag - b.imag) < epsilon;
+  }
+};
+
+template <typename OutT>
+struct EqualFunctor<phi::dtype::complex<double>, OutT> {
+  HOSTDEVICE OutT operator()(const phi::dtype::complex<double>& a,
+                             const phi::dtype::complex<double>& b) const {
+    if (isinf(a.real) || isinf(a.imag) || isinf(b.real) || isinf(b.imag)) {
+      return a == b;
+    }
+    if (isnan(a.real) || isnan(a.imag) || isnan(b.real) || isnan(b.imag)) {
+      return false;
+    }
+    double epsilon = 1e-8;
+    return std::abs(a.real - b.real) < epsilon &&
+           std::abs(a.imag - b.imag) < epsilon;
+  }
+};


这俩能合并为以下代码吗（可以参考: Paddle/paddle/phi/kernels/funcs/matrix_inverse.h）？

Suggested change

template <typename OutT>

struct EqualFunctor<phi::dtype::complex<float>, OutT> {

HOSTDEVICE OutT operator()(const phi::dtype::complex<float>& a,

const phi::dtype::complex<float>& b) const {

if (isinf(a.real) || isinf(a.imag) || isinf(b.real) || isinf(b.imag)) {

return a == b;

}

if (isnan(a.real) || isnan(a.imag) || isnan(b.real) || isnan(b.imag)) {

return false;

}

float epsilon = 1e-8f;

return std::abs(a.real - b.real) < epsilon &&

std::abs(a.imag - b.imag) < epsilon;

}

};

template <typename OutT>

struct EqualFunctor<phi::dtype::complex<double>, OutT> {

HOSTDEVICE OutT operator()(const phi::dtype::complex<double>& a,

const phi::dtype::complex<double>& b) const {

if (isinf(a.real) || isinf(a.imag) || isinf(b.real) || isinf(b.imag)) {

return a == b;

}

if (isnan(a.real) || isnan(a.imag) || isnan(b.real) || isnan(b.imag)) {

return false;

}

double epsilon = 1e-8;

return std::abs(a.real - b.real) < epsilon &&

std::abs(a.imag - b.imag) < epsilon;

}

};

template <typename InT, typename OutT>

struct EqualFunctor<phi::dtype::complex<InT>, OutT> {

HOSTDEVICE OutT operator()(const phi::dtype::complex<InT>& a,

const phi::dtype::complex<InT>& b) const {

if (isinf(a.real) || isinf(a.imag) || isinf(b.real) || isinf(b.imag)) {

return a == b;

}

if (isnan(a.real) || isnan(a.imag) || isnan(b.real) || isnan(b.imag)) {

return false;

}

InT epsilon = 1e-8;

return std::abs(a.real - b.real) < epsilon &&

std::abs(a.imag - b.imag) < epsilon;

}

};

原来还能这么用!

HydrogenSulfate · 2024-12-05T09:47:39Z

@MrXnneHang 包的编译好像有点问题，可以本地测试下编译和安装

MrXnneHang · 2024-12-05T09:50:04Z

我正在给我家里的电脑配环境，我笔记本一直都是编译cpu版本...

HydrogenSulfate · 2024-12-05T10:02:05Z

paddle/phi/kernels/funcs/compare_functors.h

+struct EqualFunctor<phi::dtype::complex<T>, OutT> {
+  HOSTDEVICE OutT operator()(const phi::dtype::complex<float>& a,
+                             const phi::dtype::complex<float>& b) const {
+    if (isinf(a.real) || isinf(a.imag) || isinf(b.real) || isinf(b.imag)) {
+      return a == b;
+    }
+    if (isnan(a.real) || isnan(a.imag) || isnan(b.real) || isnan(b.imag)) {
+      return false;
+    }
+    float epsilon = 1e-8f;
+    return std::abs(a.real - b.real) < epsilon &&
+           std::abs(a.imag - b.imag) < epsilon;
+  }


我对这里的isnan有疑惑，cpu可以用std::isnan，而GPU应该使用isnan，这两者能通用吗？参考代码：Paddle/paddle/phi/kernels/impl/isfinite_kernel_impl.h

可能isnan通用而std::isnan只能用于cpu?
我注意到原本的EqualFunctor在cpu和gpu上的注册方式完全一致。并且也只采用了isnan。
我编译gpu版本的时候都换成std::isnan测试下看看

可能isnan通用而std::isnan只能用于cpu? 我注意到原本的EqualFunctor在cpu和gpu上的注册方式完全一致。并且也只采用了isnan。我编译gpu版本的时候都换成std::isnan测试下看看

如果不通用可以这样：

全部改用std::isnan似乎编译gpu版本报错方法在deviece上未实现.
全部用isnancpu版本不会报错功能正常，但是好像gpu版本好像不走这些Functors.

我有在模板里面printf一些信息，但是cpu版本显示而gpu版本里面不显示...
分别测的是complexTensor==complexTensor和floatTensor==floatTensor

全部改用std::isnan似乎编译gpu版本报错方法在deviece上未实现. 全部用isnancpu版本不会报错功能正常，但是好像gpu版本好像不走这些Functors. 我有在模板里面printf一些信息，但是cpu版本显示而gpu版本里面不显示... 分别测的是complexTensor==complexTensor和floatTensor==floatTensor

GPU上调用的代码不能是std::系列的，参考: Paddle/paddle/phi/kernels/impl/isfinite_kernel_impl.h

目前我全部使用isnan。
但是我找不到equal在哪里注册的<GPU,AllLayout,int,float....>那些。
我注意到我在gpu版本中,对于常数的==是可以正常运行的，但是如果对于复数，会报错在未注册GPU上complex64，complex128
而且我注意到似乎gpu版本的常数==不会进入EqualFunctor主模板或者特化模板中。
参考:#69968 (comment)

…e into compare-complex

HydrogenSulfate · 2024-12-11T05:28:02Z

我验证一下想法。因为经过排查问题出在单测。因为我通过打印调试信息发现GPU版本ctest进入各个模板也都没有问题。似乎只是coverage计算的时候被排除了？最终一个都没有统计到。

好的

HydrogenSulfate · 2024-12-11T05:29:11Z

单测加一下skip，跳过xpu的设备

MrXnneHang · 2024-12-12T13:01:34Z

paddle/phi/kernels/funcs/compare_functors.h

-template <typename T>
-struct EqualFunctor<phi::dtype::complex<T>> {
-  HOSTDEVICE bool operator()(const phi::dtype::complex<T>& a,
-                             const phi::dtype::complex<T>& b) const {


特化模板应该期望得到引用或者是一个指向变量的指针。
而我五天前因为包编译错误把引用给去了。导致期望得到引用或者变量指针匹配不到传入临时变量类型，这很大概率是CI一直C++ Coverage为0的原因。
ba8cbe8

如果直接引用，会产生这个冲突:reinterpret_cast casts away qualifiers

8eb19ee

我尝试用static_cast来转，但是引用输入是不期望被修改的。编译会出错。

不能在operater中引用变量->不能使用特化模板->合并模板。

让我疑惑的是，我在本地printf可以打印出特化模板的内容。
可能的解释是下午我加入printf后出现的static check的警告:

You are using GPU version Paddle, but your CUDA device is not set properly. CPU device will be used by default.

加入printf后，GPU陷入一种薛定谔状态，它也不知道自己是CPU还是GPU。

让我疑惑的是，我在本地printf可以打印出特化模板的内容。可能的解释是下午我加入printf后出现的static check的警告:

You are using GPU version Paddle, but your CUDA device is not set properly. CPU device will be used by default.

加入printf后，GPU陷入一种薛定谔状态，它也不知道自己是CPU还是GPU。

Coverage的覆盖率测试可能是存在问题的，如果仅仅是coverage没通过，而本地自测能够打印出coverage中报告未覆盖（红色）的内容，就先不用管。

似乎确实是CI的问题
不过也确定了引用类型可以匹配上变量类型。

MrXnneHang · 2024-12-13T00:17:52Z

paddle/phi/kernels/funcs/compare_functors.h

+      }
+      if (isinf(static_cast<float>(a)) || isinf(static_cast<float>(b))) {
+        return static_cast<OutT>(a == b);
+      }


这里我把泛化模板的isnan的检查提前到isinf之前。
这样子在碰到inf==nan的情况不用多算一次a==b。结果是一样的。逻辑是只要碰到nan直接抛出false

HydrogenSulfate · 2024-12-13T12:24:33Z

@MrXnneHang 覆盖率没过

MrXnneHang · 2024-12-13T12:32:24Z

@MrXnneHang 覆盖率没过

我本地编译和测试的。

MrXnneHang · 2024-12-13T12:35:34Z

即使没有用特化模板它的覆盖率也是0,这是上次合并模板后的CI结果

25a024d

HydrogenSulfate · 2024-12-13T12:43:01Z

即使没有用特化模板它的覆盖率也是0,这是上次合并模板后的CI结果

25a024d

好的

HydrogenSulfate

有一个问题可以看一下

HydrogenSulfate · 2024-12-16T06:36:48Z

paddle/phi/kernels/funcs/compare_functors.h

-        return static_cast<OutT>(a == b);
-      if (isnan(static_cast<float>(a)) || isnan(static_cast<float>(b)))
+      if (isnan(static_cast<float>(a)) || isnan(static_cast<float>(b))) {
        return static_cast<OutT>(false);
+      }
+      if (isinf(static_cast<float>(a)) || isinf(static_cast<float>(b))) {
+        return static_cast<OutT>(a == b);


std::is_floating_point分支下，为什么要再转成float呢？

/* IsnanFunctor */ template <typename T> __global__ void IsnanCUDAKernel( const T* in_data, int num, bool* out_data, typename std::enable_if<std::is_floating_point<T>::value>::type* = 0) { unsigned int idx = threadIdx.x + blockIdx.x * blockDim.x; for (int i = idx; i < num; i += blockDim.x * gridDim.x) { const T& a = in_data[i]; out_data[i] = isnan(a); } }

可能是处于并发检查的性能和内存考虑的？
因为对于每个数都要检查一遍，如果是double的话，长度是float的两倍。而且如果都是相同的数据类型的话对于GPU本身指令特化的结构也会更有利一些。
而且isnan和isinf的话确实也没必要保持double。

我去改成float

/* IsnanFunctor */ template <typename T> __global__ void IsnanCUDAKernel( const T* in_data, int num, bool* out_data, typename std::enable_if<std::is_floating_point<T>::value>::type* = 0) { unsigned int idx = threadIdx.x + blockIdx.x * blockDim.x; for (int i = idx; i < num; i += blockDim.x * gridDim.x) { const T& a = in_data[i]; out_data[i] = isnan(a); } }

可能是处于并发检查的性能和内存考虑的？因为对于每个数都要检查一遍，如果是double的话，长度是float的两倍。而且如果都是相同的数据类型的话对于GPU本身指令特化的结构也会更有利一些。而且isnan和isinf的话确实也没必要保持double。

这会不会导致原本double的数被转为float后结果就不正确了呢

哦如果原本是这样写的话，那先保持原状吧

HydrogenSulfate · 2024-12-16T07:47:34Z

paddle/phi/kernels/funcs/compare_functors.h

+template <typename T>
+struct NotEqualFunctor<phi::dtype::complex<T>> {
+  HOSTDEVICE bool operator()(const phi::dtype::complex<T> a,
+                             const phi::dtype::complex<T> b) const {
+    return !EqualFunctor<phi::dtype::complex<T>>()(a, b);
+  }
+};


这里能不改动原本的模板参数吗？

这个我需要测试一下，我记得NotEqual没有特化也是正常工作的。

删除掉NotEqual特化模板没有影响，依然可以进Equal特化和泛化。

HydrogenSulfate

LGTM

MrXnneHang added 3 commits December 5, 2024 08:31

chore: support complex on cpu.

bd209c7

test: add unitest for complex compare.

92c474f

chore: register on gpu.

ccf4ed1

paddle-bot bot added the contributor External developers label Dec 5, 2024

luotao1 added the HappyOpenSource 快乐开源活动issue与PR label Dec 5, 2024

luotao1 assigned luotao1 and HydrogenSulfate Dec 5, 2024

HydrogenSulfate reviewed Dec 5, 2024

View reviewed changes

paddle/phi/kernels/kps/compare_kernel.cu Outdated Show resolved Hide resolved

MrXnneHang added 8 commits December 5, 2024 12:48

chore: add template for complex. split kernel by complex.

2f4f8db

fix: use phi::dtype::complex.real/imag to read content.

d755674

chore: register complex_equal、complex_not_equal func.

a251348

test: remove func complex_not_equal、complex_equal, only test `not…

6b1c86d

…_equal` and `equal`

ref: 采用特化模板实现对复数支持

30990b6

chore: register complex64,complex128 for equal、not_equal

55413c8

ref: remove complex_not_equal,complex_equal.

ef24451

ref: remove useless blank line.

b27a9ba

MrXnneHang changed the title ~~【Paddle Tensor 第二期复数类型支持问题No.7、9、10】为compare_kernel添加复数类型支持。~~ 【Paddle Tensor 第二期复数类型支持问题No.7、9、10】为not_equal、paddle.equal添加复数类型支持。 Dec 5, 2024

MrXnneHang added 2 commits December 5, 2024 16:57

chore: register on gpu.

1fb3773

chore: 恢复换行.

13a81e1

HydrogenSulfate reviewed Dec 5, 2024

View reviewed changes

MrXnneHang added 4 commits December 6, 2024 10:22

ref: combine <float> and <double> into <T>.

5f0a5fb

ref: combine <float> and <double> into <T>.

70463d4

chore: register legacy raw_kernel.

fc8beaa

Merge branch 'compare-complex-gpu-fix' of github.com:MrXnneHang/Paddl…

d643035

…e into compare-complex

MrXnneHang added 4 commits December 10, 2024 23:26

fix: unittest.main()

7670074

test: minimize test.

dde0cd6

fix: a_np == b

0f42ccb

fix: a_np == b

e2249a1

MrXnneHang added 3 commits December 11, 2024 15:55

fix: Use OpTest and set checkpir=True

50f54ab

test: add special case for y=complex

6ff012e

refactor: 更换skip方式.

6af2648

MrXnneHang force-pushed the compare-complex branch from bf5ac03 to 6af2648 Compare December 12, 2024 04:05

MrXnneHang added 2 commits December 12, 2024 12:12

test: 尝试打印调试信息.

c2ffb16

test: 使用引用而不是变量。

c5413b0

MrXnneHang force-pushed the compare-complex branch from 7be9d62 to c5413b0 Compare December 12, 2024 11:05

fix: use static_cast instead of reinterpret_cast.

8eb19ee

MrXnneHang force-pushed the compare-complex branch from b071f4e to 8eb19ee Compare December 12, 2024 12:05

合并模板,防止引用变量被修改.

25a024d

MrXnneHang commented Dec 12, 2024

View reviewed changes

恢复特化模板的写法。提前检查nan.

0a6e65e

MrXnneHang commented Dec 13, 2024

View reviewed changes

MrXnneHang mentioned this pull request Dec 15, 2024

[WeeklyReports] 2024.11.29~2024.12.13周报收集 PFCCLab/Starter#463

Open

74 tasks

HydrogenSulfate reviewed Dec 16, 2024

View reviewed changes

MrXnneHang added 2 commits December 16, 2024 15:48

pef: 使用float进行isnan的检查。

fb53669

ref: 简化代码结构.删除NotEqual特化模板.

2d6ecb0

HydrogenSulfate approved these changes Dec 17, 2024

View reviewed changes

HydrogenSulfate merged commit 1e4a234 into PaddlePaddle:develop Dec 17, 2024
28 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

【Paddle Tensor 第二期复数类型支持问题No.9、10】为`paddle.not_equal`、`paddle.equal`添加复数类型支持。 #69968

【Paddle Tensor 第二期复数类型支持问题No.9、10】为`paddle.not_equal`、`paddle.equal`添加复数类型支持。 #69968

MrXnneHang commented Dec 5, 2024 •

edited

Loading

paddle-bot bot commented Dec 5, 2024

HydrogenSulfate left a comment

HydrogenSulfate left a comment

HydrogenSulfate Dec 5, 2024

MrXnneHang Dec 5, 2024 •

edited

Loading

HydrogenSulfate commented Dec 5, 2024

MrXnneHang commented Dec 5, 2024

HydrogenSulfate Dec 5, 2024

MrXnneHang Dec 5, 2024

HydrogenSulfate Dec 5, 2024

MrXnneHang Dec 9, 2024 •

edited

Loading

HydrogenSulfate Dec 9, 2024

MrXnneHang Dec 9, 2024 •

edited

Loading

HydrogenSulfate commented Dec 11, 2024

HydrogenSulfate commented Dec 11, 2024

MrXnneHang Dec 12, 2024 •

edited

Loading

MrXnneHang Dec 12, 2024 •

edited

Loading

HydrogenSulfate Dec 12, 2024 •

edited

Loading

MrXnneHang Dec 13, 2024 •

edited

Loading

MrXnneHang Dec 13, 2024 •

edited

Loading

HydrogenSulfate commented Dec 13, 2024 •

edited

Loading

MrXnneHang commented Dec 13, 2024

MrXnneHang commented Dec 13, 2024

HydrogenSulfate commented Dec 13, 2024

HydrogenSulfate left a comment •

edited

Loading

HydrogenSulfate Dec 16, 2024

MrXnneHang Dec 16, 2024

MrXnneHang Dec 16, 2024

HydrogenSulfate Dec 16, 2024

HydrogenSulfate Dec 16, 2024

HydrogenSulfate Dec 16, 2024

MrXnneHang Dec 16, 2024

MrXnneHang Dec 16, 2024

HydrogenSulfate left a comment

【Paddle Tensor 第二期 复数类型支持问题No.9、10】为paddle.not_equal、paddle.equal添加复数类型支持。 #69968

【Paddle Tensor 第二期 复数类型支持问题No.9、10】为paddle.not_equal、paddle.equal添加复数类型支持。 #69968

Conversation

MrXnneHang commented Dec 5, 2024 • edited Loading

PR Category

PR Types

Description

error 1 (solved)

error 2 (solved)

error 3 (solved) :

paddle-bot bot commented Dec 5, 2024

HydrogenSulfate left a comment

Choose a reason for hiding this comment

HydrogenSulfate left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

MrXnneHang Dec 5, 2024 • edited Loading

Choose a reason for hiding this comment

HydrogenSulfate commented Dec 5, 2024

MrXnneHang commented Dec 5, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

MrXnneHang Dec 9, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

MrXnneHang Dec 9, 2024 • edited Loading

Choose a reason for hiding this comment

HydrogenSulfate commented Dec 11, 2024

HydrogenSulfate commented Dec 11, 2024

MrXnneHang Dec 12, 2024 • edited Loading

Choose a reason for hiding this comment

MrXnneHang Dec 12, 2024 • edited Loading

Choose a reason for hiding this comment

HydrogenSulfate Dec 12, 2024 • edited Loading

Choose a reason for hiding this comment

MrXnneHang Dec 13, 2024 • edited Loading

Choose a reason for hiding this comment

MrXnneHang Dec 13, 2024 • edited Loading

Choose a reason for hiding this comment

HydrogenSulfate commented Dec 13, 2024 • edited Loading

MrXnneHang commented Dec 13, 2024

MrXnneHang commented Dec 13, 2024

HydrogenSulfate commented Dec 13, 2024

HydrogenSulfate left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

HydrogenSulfate left a comment

Choose a reason for hiding this comment

【Paddle Tensor 第二期复数类型支持问题No.9、10】为`paddle.not_equal`、`paddle.equal`添加复数类型支持。 #69968

【Paddle Tensor 第二期复数类型支持问题No.9、10】为`paddle.not_equal`、`paddle.equal`添加复数类型支持。 #69968

MrXnneHang commented Dec 5, 2024 •

edited

Loading

MrXnneHang Dec 5, 2024 •

edited

Loading

MrXnneHang Dec 9, 2024 •

edited

Loading

MrXnneHang Dec 9, 2024 •

edited

Loading

MrXnneHang Dec 12, 2024 •

edited

Loading

MrXnneHang Dec 12, 2024 •

edited

Loading

HydrogenSulfate Dec 12, 2024 •

edited

Loading

MrXnneHang Dec 13, 2024 •

edited

Loading

MrXnneHang Dec 13, 2024 •

edited

Loading

HydrogenSulfate commented Dec 13, 2024 •

edited

Loading

HydrogenSulfate left a comment •

edited

Loading