Add Split-18: new num_outputs attribute #4358
Conversation
onnx/defs/tensor/defs.cc
Outdated
| @@ -682,7 +688,7 @@ ONNX_OPERATOR_SET_SCHEMA( | |||
| ")"); | |||
| } | |||
| } else { // no value available for 'split' | |||
| int num_outputs = static_cast<int>(ctx.getNumOutputs()); | |||
| int num_outputs = static_cast<int>(getAttribute(ctx, "num_outputs", 0)); | |||
There was a problem hiding this comment.
Would be good to check that the attribute is specified, and produce an error message otherwise. Similarly, it may be useful to check in the then-branch that the attribute is not specified.
There was a problem hiding this comment.
Done, please let me know if my changes address your comment.
b4cb6f7 to
88df4af
Compare
Signed-off-by: p-wysocki <przemyslaw.wysocki@intel.com>
9bb4825 to
0702fab
Compare
Signed-off-by: p-wysocki <przemyslaw.wysocki@intel.com>
| set ONNX_ML=0 | ||
|
|
||
| # UNIX | ||
| # export ONNX_ML=0 |
There was a problem hiding this comment.
Is that # in the second line a typo or intentional?
There was a problem hiding this comment.
It was a result of my failed attempt at fixing non-signed-off commit issue. My git got messed up, I'm going to open a new, clean PR. Sorry for inconvenience.
|
Do you need to rebase? Or, something is off, it is showing other recently merged PRs in the diff here, which is a bit confusing. |
| split.reserve(ctx.getNumOutputs()); | ||
| for (int i = 0; i < static_cast<int>(ctx.getNumOutputs()); i++) { | ||
| split.push_back(chunk_size); | ||
| const auto num_outputs_attr = ctx.getAttribute("num_outputs"); |
There was a problem hiding this comment.
Suggest moving the above line out of the else-branch to before the if-statement, and adding a check in the then-branch that num_outputs_attr == nullptr (to catch the case where both are specified, which is also erroneous).
There was a problem hiding this comment.
I added a shape inference error in case both are specified. The PR has been moved to #4389.
I had an issue with a non-signed off commit, and then broke something trying to fix it. I'll get to fixing it today. |
Signed-off-by: p-wysocki <przemyslaw.wysocki@intel.com>
|
I moved the PR to #4389, my attempts at fixing non-signed-off commits led me to breaking my git, creating a new PR was a quicker and less messy solution. |
Signed-off-by: Przemyslaw Wysocki przemyslaw.wysocki@intel.com
Description
num_outputsdetermining the number of equally sized output tensors.Motivation and Context