Skip to content

[Accuracy diff No.119、145] Fix accuracy diff for paddle.nn.functional.gaussian_nll_loss、paddle.Tensor.fill_diagonal_tensor API #272

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

Merged
merged 4 commits into from
Jun 17, 2025
Merged

[Accuracy diff No.119、145] Fix accuracy diff for paddle.nn.functional.gaussian_nll_loss、paddle.Tensor.fill_diagonal_tensor API #272

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
30448b0
modify grads and dtype diff
cangtianhuang Jun 16, 2025
1417478
copy configs
cangtianhuang Jun 16, 2025
4a0f1aa
Merge remote-tracking branch 'upstream/main' into develop-fast
cangtianhuang Jun 17, 2025
34de24b
refine outputs and grads special disposal
cangtianhuang Jun 17, 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
66 changes: 33 additions & 33 deletions tester/accuracy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ def __init__(self, api_config, test_amp):
self.test_amp = test_amp
self.converter = get_converter()

#@func_set_timeout(600)
# @func_set_timeout(600)
def test(self):
if self.need_skip():
print("[Skip]", flush=True)
Expand Down Expand Up @@ -219,17 +219,23 @@ def test(self):
write_to_log("paddle_error", self.api_config.config)
return

if self.api_config.api_name == "paddle.incubate.nn.functional.fused_rms_norm":
if self.api_config.api_name == "paddle.incubate.nn.functional.fused_rms_norm":
paddle_output = paddle_output[0]
if self.api_config.api_name == "paddle.unique":
elif self.api_config.api_name == "paddle.unique":
if "return_index=True" in self.api_config.config:
paddle_output = list(paddle_output)
paddle_output.pop(1)
paddle_output = tuple(paddle_output)
if self.api_config.api_name in {"paddle.mode", "paddle.Tensor.mode"}:
elif self.api_config.api_name in {
"paddle.mode",
"paddle.Tensor.mode",
}:
paddle_output = paddle_output[0]
torch_output = torch_output[0]
if self.api_config.api_name in {"paddle.strided_slice", "paddle.vander"} and any(s < 0 for s in paddle_output.strides):
elif self.api_config.api_name in {
"paddle.strided_slice",
"paddle.vander",
} and any(s < 0 for s in paddle_output.strides):
# torch's from_dlpack now don't support negative strides
paddle_output = paddle_output.contiguous()

Expand Down Expand Up @@ -360,38 +366,32 @@ def test(self):
if self.api_config.api_name == "paddle.Tensor.__setitem__":
torch_out_grads = torch_out_grads[0]
paddle_out_grads = paddle_out_grads[0]

# All configs that not compared with torch
# should be moved to tester/api_config/5_accuracy/grads_diff.txt
if self.api_config.api_name == "paddle.tensordot":
paddle_out_grads = paddle_out_grads[:2]
if self.api_config.api_name == "paddle.combinations":
paddle_out_grads = []
torch_out_grads = []
if self.api_config.api_name == "paddle.diagonal_scatter":
paddle_out_grads = paddle_out_grads[:1]
torch_out_grads = torch_out_grads[:1]
if self.api_config.api_name == "paddle.lerp":
paddle_out_grads = paddle_out_grads[:2]
torch_out_grads = torch_out_grads[:2]
if self.api_config.api_name == "paddle.nn.utils.parameters_to_vector":
paddle_out_grads = []
torch_out_grads = []
if self.api_config.api_name == "paddle.nn.functional.kl_div":
paddle_out_grads = paddle_out_grads[:1]
torch_out_grads = torch_out_grads[:1]
if self.api_config.api_name == "paddle.scale":
paddle_out_grads = paddle_out_grads[0]
torch_out_grads = torch_out_grads[0]
if self.api_config.api_name == "paddle.nn.functional.gaussian_nll_loss":

# All configs that not compared with torch should be copied
# to tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt
if self.api_config.api_name in {
"paddle.lerp",
"paddle.tensordot",
}:
paddle_out_grads = paddle_out_grads[:2]
torch_out_grads = torch_out_grads[:2]
if self.api_config.api_name == "paddle.nn.functional.binary_cross_entropy":
paddle_out_grads = paddle_out_grads[0]
torch_out_grads = torch_out_grads[0]
if self.api_config.api_name == "paddle.nn.functional.binary_cross_entropy_with_logits":
elif self.api_config.api_name in {
"paddle.Tensor.fill_diagonal_tensor",
"paddle.diagonal_scatter",
"paddle.nn.functional.binary_cross_entropy",
"paddle.nn.functional.binary_cross_entropy_with_logits",
"paddle.nn.functional.gaussian_nll_loss",
"paddle.nn.functional.kl_div",
"paddle.scale",
}:
paddle_out_grads = paddle_out_grads[0]
torch_out_grads = torch_out_grads[0]
elif self.api_config.api_name in {
"paddle.combinations",
"paddle.nn.utils.parameters_to_vector",
}:
paddle_out_grads = []
torch_out_grads = []

if isinstance(paddle_out_grads, paddle.Tensor):
if isinstance(torch_out_grads, torch.Tensor):
Expand Down
34 changes: 34 additions & 0 deletions tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,6 +54,40 @@ paddle.scale(Tensor([2, 1, 1, 2, 3],"float64"), scale=Tensor([],"float32"), )
paddle.scale(Tensor([2, 1, 2, 3],"float32"), scale=Tensor([],"float32"), )
paddle.scale(Tensor([2, 1, 2, 3],"float64"), scale=Tensor([],"float32"), )
paddle.scale(Tensor([2, 3, 4, 5, 6],"float32"), scale=Tensor([1],"float32"), )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2, 3],"float32"), Tensor([10, 2, 3],"float32"), Tensor([10, 2, 1],"float32"), False, 1e-06, "none", None, )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2, 3],"float32"), Tensor([10, 2, 3],"float32"), Tensor([10, 2, 1],"float32"), full=False, reduction="none", )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2, 3],"float32"), Tensor([10, 2, 3],"float32"), Tensor([10, 2],"float32"), False, 1e-06, "none", None, )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2, 3],"float32"), Tensor([10, 2, 3],"float32"), Tensor([10, 2],"float32"), full=False, reduction="none", )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), False, 1e-06, "none", None, )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), True, 1e-06, "mean", None, )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), True, 1e-06, "sum", None, )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), full=False, reduction="none", )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), full=True, reduction="mean", )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), Tensor([10, 2],"float32"), full=True, reduction="sum", )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float64"), Tensor([10, 2],"float64"), Tensor([10, 2],"float64"), False, 1e-06, "none", None, )
paddle.nn.functional.gaussian_nll_loss(Tensor([10, 2],"float64"), Tensor([10, 2],"float64"), Tensor([10, 2],"float64"), full=False, reduction="none", )
paddle.Tensor.fill_diagonal_tensor(Tensor([2, 4, 3, 2],"float32"), Tensor([2, 2, 3],"float32"), offset=0, dim1=1, dim2=2, )
paddle.Tensor.fill_diagonal_tensor(Tensor([2, 4, 3, 2],"float64"), Tensor([2, 2, 3],"float64"), offset=0, dim1=1, dim2=2, )
paddle.Tensor.fill_diagonal_tensor(Tensor([2, 4, 4],"float32"), Tensor([4, 2],"float32"), 0, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([2, 4, 4],"float64"), Tensor([4, 2],"float64"), 0, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([2, 4, 7],"float32"), Tensor([2, 4],"float32"), 0, 1, 2, )
paddle.Tensor.fill_diagonal_tensor(Tensor([2, 4, 7],"float64"), Tensor([2, 4],"float64"), 0, 1, 2, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float32"), Tensor([1],"float32"), -2, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float32"), Tensor([2],"float32"), -1, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float32"), Tensor([2],"float32"), 1, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float32"), Tensor([3],"float32"), 0, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float64"), Tensor([1],"float64"), -2, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float64"), Tensor([2],"float64"), -1, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float64"), Tensor([2],"float64"), 1, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([3, 3],"float64"), Tensor([3],"float64"), 0, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([4, 3],"float32"), Tensor([2],"float32"), offset=1, dim1=0, dim2=1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([4, 3],"float32"), Tensor([3],"float32"), offset=-1, dim1=0, dim2=1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([4, 3],"float32"), Tensor([3],"float32"), offset=0, dim1=0, dim2=1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([4, 3],"float64"), Tensor([2],"float64"), offset=1, dim1=0, dim2=1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([4, 3],"float64"), Tensor([3],"float64"), offset=-1, dim1=0, dim2=1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([4, 3],"float64"), Tensor([3],"float64"), offset=0, dim1=0, dim2=1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([5, 3],"float32"), Tensor([3],"float32"), 0, 0, 1, )
paddle.Tensor.fill_diagonal_tensor(Tensor([5, 3],"float64"), Tensor([3],"float64"), 0, 0, 1, )
paddle.nn.functional.binary_cross_entropy(Tensor([1, 1, 2],"float64"), label=Tensor([1, 1, 2],"float64"), weight=None, reduction="mean", name=None, )
paddle.nn.functional.binary_cross_entropy(Tensor([100, 1],"float32"), Tensor([100, 1],"float32"), reduction="none", )
paddle.nn.functional.binary_cross_entropy(Tensor([100, 80],"float32"), Tensor([100, 80],"float32"), reduction="none", )
Expand Down
4 changes: 2 additions & 2 deletions tester/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,8 +117,8 @@
]
)

# All configs that report dtype diff when not in not_check_dtype list
# should be moved to tester/api_config/5_accuracy/dtype_diff.txt
# All configs that report dtype diff when not in not_check_dtype list should be
# copied to tester/api_config/5_accuracy/accuracy_gpu_error_dtype_diff.txt
not_check_dtype = frozenset(
[
"paddle.where",
Expand Down