Skip to content

[Accuracy diff No.15] Fix accuracy diff for cumprod API #72897

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 8 commits into from
Closed

[Accuracy diff No.15] Fix accuracy diff for cumprod API #72897

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
706ac0c
[Accuracy diff No.15] Fix accuracy diff for cumprod API
ooooo-create May 22, 2025
68c3645
refine
ooooo-create May 24, 2025
a6a2f67
fix typos
ooooo-create May 27, 2025
7960404
apply review
ooooo-create May 29, 2025
0792f72
refine
ooooo-create May 31, 2025
c04c965
try
ooooo-create May 31, 2025
7ed04a4
try
ooooo-create May 31, 2025
9ade7fb
Merge branch 'develop' into accuray_diff_cumprod
ooooo-create Jun 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 16 additions & 2 deletions python/paddle/tensor/math.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4722,8 +4722,22 @@ def cumprod(

"""

if dtype is not None and x.dtype != convert_np_dtype_to_dtype_(dtype):
x = cast(x, dtype)
if dtype is not None:
target_dtype = convert_np_dtype_to_dtype_(dtype)
if x.dtype != target_dtype:
x = cast(x, target_dtype)
else:
converted_x_dtype = convert_dtype(x.dtype)
# use the default platform integer when integer dtype with a precision less than that of the default platform integer
if converted_x_dtype in {
"bool",
"uint16",
"int8",
"int16",
"int32",
"uint8",
}:
Copy link
Contributor

@lshpku lshpku May 29, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这段代码涉及几次dict查找，比较耗时，过不了API性能监控（这个CI查得很严，python性能也要管），建议你可以在文件的适当地方定义一个纯DataType的set，像这样：

_supported_int_like_types = {
    DataType.BOOL,
    DataType.INT8,
    DataType.INT16,
    DataType.INT32,
    DataType.INT64,
    DataType.UINT8
}

然后在这里直接判断：

if x.dtype in _supported_int_like_types:
    x = cast(x, "int64")

当然能不能过CI还要你自己试一下

Copy link
Contributor Author

@ooooo-create ooooo-create May 29, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks!我试一下~

Copy link
Contributor Author

@ooooo-create ooooo-create May 29, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这样改了发现还是不能过耶，其实我们也可以通过手动设置 dtype 来达到 paddle 和 torch api 之间的切换，这样看起来也不算是精度问题吧，属于 api 设计的考量？目前可能去 paddleAPITest 仓库进行修改更加合适～

Copy link
Contributor Author

@ooooo-create ooooo-create May 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

提交了 PaddleAPITest 的修改 pr : PFCCLab/PaddleAPITest#212

x = cast(x, "int64")

if in_dynamic_or_pir_mode():
return _C_ops.cumprod(x, dim, False, False)
Expand Down
34 changes: 30 additions & 4 deletions test/legacy_test/test_cumprod_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -261,6 +261,7 @@ def init_dtype(self):

def setUp(self):
paddle.enable_static()
self.target_dtype = None
self.init_dtype()
self.x = (np.random.rand(2, 3, 10, 10) + 0.5).astype(self.dtype)
self.place = []
Expand All @@ -281,10 +282,10 @@ def test_static_api(self):
def run(place):
with paddle.static.program_guard(paddle.static.Program()):
x = paddle.static.data('X', self.shape, dtype=self.dtype)
out = paddle.cumprod(x, -2)
out = paddle.cumprod(x, -2, self.target_dtype)
exe = paddle.static.Executor(place)
res = exe.run(feed={'X': self.x}, fetch_list=[out])
out_ref = np.cumprod(self.x, -2)
out_ref = np.cumprod(self.x, -2, self.target_dtype)

for r in res:
np.testing.assert_allclose(out_ref, r, rtol=1e-05)
Expand All @@ -297,15 +298,40 @@ def test_dygraph_api(self):
def run(place):
paddle.disable_static(place)
x = paddle.to_tensor(self.x)
out = paddle.cumprod(x, 1)
out_ref = np.cumprod(self.x, 1)
out = paddle.cumprod(x, 1, self.target_dtype)
out_ref = np.cumprod(self.x, 1, self.target_dtype)
np.testing.assert_allclose(out_ref, out.numpy(), rtol=1e-05)
paddle.enable_static()

for place in self.place:
run(place)


class TestCumprodAPICase1(TestCumprodAPI):
def init_dtype(self):
self.dtype = 'int32'
self.shape = [2, 3, 10, 10]


class TestCumprodAPICase2(TestCumprodAPI):
def init_dtype(self):
self.dtype = 'bool'
self.shape = [2, 3, 10, 10]


class TestCumprodAPICase3(TestCumprodAPI):
def init_dtype(self):
self.dtype = 'int64'
self.shape = [2, 3, 10, 10]


class TestCumprodAPICase4(TestCumprodAPI):
def init_dtype(self):
self.dtype = 'float32'
self.shape = [2, 3, 10, 10]
self.target_dtype = 'float64'


# test function.
class TestCumprodReverse(TestCumprod):
def init_dtype(self):
Expand Down
Loading