Closed orausch closed 2 years ago
Merging #95 (a3488aa) into master (ddd1120) will increase coverage by
7.32%
. The diff coverage is93.55%
.
@@ Coverage Diff @@
## master #95 +/- ##
==========================================
+ Coverage 60.07% 67.39% +7.32%
==========================================
Files 58 58
Lines 6870 6955 +85
==========================================
+ Hits 4127 4687 +560
+ Misses 2743 2268 -475
Impacted Files | Coverage Δ | |
---|---|---|
...ml/onnx/op_implementations/pure_implementations.py | 69.60% <85.71%> (+4.39%) |
:arrow_up: |
daceml/onnx/environments/onnxruntime.py | 87.63% <93.10%> (+18.88%) |
:arrow_up: |
daceml/onnx/onnx_importer.py | 92.28% <96.88%> (-2.79%) |
:arrow_down: |
daceml/onnx/__init__.py | 100.00% <100.00%> (ø) |
|
daceml/onnx/nodes/node_codegen.py | 89.27% <100.00%> (+17.21%) |
:arrow_up: |
.../onnx/op_implementations/img_op_implementations.py | 66.79% <100.00%> (-0.37%) |
:arrow_down: |
daceml/onnx/op_implementations/utils.py | 100.00% <100.00%> (+4.76%) |
:arrow_up: |
daceml/onnx/schema.py | 95.62% <100.00%> (ø) |
|
daceml/onnx/shape_inference/shape_inference.py | 100.00% <100.00%> (ø) |
|
daceml/ort_api/raw_api_bindings.py | 95.58% <100.00%> (ø) |
|
... and 24 more |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update ddd1120...a3488aa. Read the comment docs.
This PR makes the ORT dependency soft. If ORT is not installed, ORT node expansion will fail. There is a new CI branch that runs tests without ORT.
Since our constant folding depends on ORT, we add more cleanup to the ONNX Importer. It now uses
onnxsim
to do some preliminary cleanup, doing a lot of the heavy lifting that our CF would do.To keep tests from failing when ORT is missing, we also add some new pure op implementations. This includes improvements to the
python_pure_op_implementation
decorator to make these more compact. In a later PR, I will apply these improvements to other pure impls (I expect to be able to simplify at least 11 implementations).