Skip to content

Support aten::div.Tensor_mode #980

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

Jump to bottom
Closed
titaiwangms opened this issue Aug 5, 2023 · 1 comment · Fixed by #988
Closed

Support aten::div.Tensor_mode #980

titaiwangms opened this issue Aug 5, 2023 · 1 comment · Fixed by #988
Assignees
@xiaowuhu @fatcat-z
Labels
contribution welcome We welcome code contributions for this module: torchlib Related to the torch/aten function lib in development

Comments

@titaiwangms
Copy link
Contributor

Essentially, rounding_mode attribute support.
I think this op blocks multiple models from torchbench.

cc @BowenBao @xiaowuhu @justinchuby
@titaiwangms titaiwangms added the module: torchlib Related to the torch/aten function lib in development label Aug 5, 2023
@titaiwangms
Copy link
Contributor Author

Feel free to close it and categorize this issue under others, if there is any.

@titaiwangms titaiwangms mentioned this issue Aug 5, 2023
Closed
@justinchuby justinchuby mentioned this issue Aug 5, 2023
Open
@justinchuby justinchuby added the contribution welcome We welcome code contributions for this label Aug 5, 2023
@justinchuby justinchuby mentioned this issue Aug 8, 2023
Merged
justinchuby added a commit that referenced this issue Aug 8, 2023
@justinchuby
Implement aten::div.Tensor_mode | feat(torchlib) (#988)
ba255f7 
`aten::div.Tensor_mode` is implemented with two ONNX functions. When
`rounding_mode` is `None`, we use `aten_div`. When it is not None, we
use `aten_div_mode`. This way we don't need to handle when
`rounding_mode==None` in `aten_div`.

For `float16` inputs we need to cast to float32 to preserve precision.
Otherwise `-inf` sometimes becomes `inf` in the output.

- Additionally registers aliases "aten::divide", "aten::true_divide" to
`aten_div`.
- Supports saving mismatches to error reports
- xfail and documents off-by-one errors with float16
(#990,
#989)

Fixes #980
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@xiaowuhu xiaowuhu

@fatcat-z fatcat-z

Labels
contribution welcome We welcome code contributions for this module: torchlib Related to the torch/aten function lib in development
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Implement aten::div.Tensor_mode | feat(torchlib) microsoft/onnxscript
4 participants
@justinchuby @titaiwangms @xiaowuhu @fatcat-z