Skip to content

[torchlib] prims::sum #2778

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
justinchuby merged 3 commits into from
Jan 22, 2026
Merged

[torchlib] prims::sum #2778

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f6b853c
[torchlib] prims::sum
justinchuby Jan 22, 2026
ac16cdf
lint
justinchuby Jan 22, 2026
233084a
test
justinchuby Jan 22, 2026
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
8 changes: 7 additions & 1 deletion onnxscript/function_libs/torch_lib/ops/prims.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -772,12 +772,18 @@ def prims_sub(self: TTensor, other: TTensor) -> TTensor:
return op.Sub(self, other)


@torch_op("prims::sum", trace_only=True)
def prims_sum(
inp: TensorType, dims: Optional[Sequence[int]], output_dtype: Optional[int] = None
) -> TensorType:
"""sum(Tensor inp, int[]? dims, *, ScalarType? output_dtype=None) -> Tensor"""

Copy link

Copilot AI Jan 22, 2026

Choose a reason for hiding this comment

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

The dims parameter should be normalized to None when it's an empty sequence, following the pattern used in prims_var (lines 839-841). ONNX's ReduceSum operator may not handle empty sequences correctly, and converting empty sequences to None ensures the operator reduces over all dimensions as expected.

Suggested change
if dims is not None and len(dims) == 0:
dims = None

Copilot uses AI. Check for mistakes.
raise NotImplementedError()
result = op.ReduceSum(inp, dims, keepdims=False)

if output_dtype is not None and output_dtype != -1:
result = op.Cast(result, to=output_dtype)

return result


def prims_svd(A: TensorType, full_matrices: bool) -> tuple[TensorType, TensorType, TensorType]:
Expand Down
12 changes: 12 additions & 0 deletions tests/function_libs/torch_lib/e2e_ops_tests.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -902,6 +902,18 @@ def forward(self, x, index, update):
)
_testing.assert_onnx_program(onnx_program)

def test_std_mean(self):
"""Test torch.std_mean which will be decomposed into prims.sum."""

class Model(torch.nn.Module):
def forward(self, x):
return torch.std_mean(x)

onnx_program = torch.onnx.export(
Model(), (torch.rand(10, 10, 10),), dynamo=True, verbose=False
)
_testing.assert_onnx_program(onnx_program)


if __name__ == "__main__":
unittest.main()
Loading