Merge 'initializers' into 'TorchScriptGraph' #480
Conversation
Codecov Report
@@ Coverage Diff @@
## main #480 +/- ##
=======================================
Coverage 72.45% 72.46%
=======================================
Files 109 109
Lines 10602 10607 +5
Branches 1093 1093
=======================================
+ Hits 7682 7686 +4
- Misses 2613 2614 +1
Partials 307 307
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. |
|
I think torch-nightly is broken by #460 |
|
@titaiwangms thx, hard to figure out what are known failures... Does the CI here depend on code from torch exporter? This api change requires modification from both sides, I was planning to merge this first hoping CI could be all green. |
I don't think onnx-script relies on torch exporter, but what part in this PR needs torch exporter modification? As far as I know, the experimental branch is broken due to lack of maintain on ort-experimental branch, and torch-nightly fails because of squeeze op. There are other two erroring with ORT that I don't quite understand, but this PR seems to be unrelated... |
|
The api changed, |
|
Fixes #389 |
|
|
||
| @property | ||
| def torch_graph(self): | ||
| return self._torch_graph | ||
|
|
||
| @property | ||
| def initializers(self) -> Mapping[str, torch.Tensor]: |
There was a problem hiding this comment.
Would a brief docstring for this property help?
| @@ -17,23 +17,22 @@ | |||
| @unittest.skipIf(version_utils.torch_older_than("2.0"), "torchscript in 1.13 not supported") | |||
| class TestTorchScriptTracingEvaluator(unittest.TestCase): | |||
| def setUp(self): | |||
| # FIXME: Currently this must match with the import line | |||
| # `from onnxscript import opset17 as op`, which restricts opset to be 17 in these | |||
There was a problem hiding this comment.
The model opset can actually be different from the local functions' opsets I think. Do you know if that's true?
There was a problem hiding this comment.
Yea it could be different under constraint.
https://github.com/onnx/onnx/blob/4b2d50334914621835cc1e8dadd4fe82b6b9876c/onnx/onnx.in.proto#L824-L828
// The operator sets imported by FunctionProto should be compatible with the ones
// imported by ModelProto. Example, if same operator set say 'A' is imported by FunctionProto
// and ModelProto then versions for the operator set may be different but,
// the operator schema returned for op_type, domain, version combination
// for both the versions should be same.
Reduce friction.