Issues

New issue

Broken CIs
#539 · justinchuby opened on Mar 20, 2023
1
[torchlib] Inefficient implementations
#2489 · justinchuby opened on Aug 14, 2025
10

Always optimize back-to-back reshapes
microsoft/onnxscript#3059 · opened on Sep 30, 2026
Always optimize back-to-back transposes
microsoft/onnxscript#3058 · opened on Sep 30, 2026
constant folding does not fold SplitToSequence layer w/o a split input
microsoft/onnxscript#3057 · opened on Sep 30, 2026
@script() with no explicit opset silently exports functions under a placeholder domain (this)
microsoft/onnxscript#3044 · opened on Sep 12, 2026
Constant folding uses onnx's plain reference ops, never onnx.reference.ops_optimized
microsoft/onnxscript#3003 · opened on Aug 19, 2026
3
check_model in _framework_apis is a no-op — torch.onnx.export(verify=True) reports success without validating anything
microsoft/onnxscript#2981 · opened on Jul 29, 2026
4
BatchNormalization layer with training_mode=1 results in incorrect ONNX model
microsoft/onnxscript#2969 · opened on Jul 22, 2026
4
Follow up on bincount + index_put
microsoft/onnxscript#2965 · opened on Jul 15, 2026
Invalid reshape after FoldConstantsPass in specific {Shape->Gather->Concat->Reshape} pattern
microsoft/onnxscript#2963 · opened on Jul 14, 2026
1
Version converter does not apply applicable conversion
microsoft/onnxscript#2938 · opened on Jun 16, 2026
1
Existing project using v0.5.x fails due to functions in the model
microsoft/onnxscript#2933 · opened on Jun 12, 2026
1
aten::conv3d onnx export bug when bias is missing
microsoft/onnxscript#2931 · opened on Jun 10, 2026
2
CI : Pin github actions to full commit SHA
microsoft/onnxscript#2903 · opened on Apr 30, 2026
6
Constant folding override
microsoft/onnxscript#2890 · opened on Apr 15, 2026
3
Verbose setting for RewritePass for better logging and debugging
microsoft/onnxscript#2876 · opened on Mar 31, 2026
2
Should we turn trace_only default to True in torchlib?
microsoft/onnxscript#2812 · opened on Feb 17, 2026
1
[Optimizer] Fold Shape -> {Slice, Gather} -> Concat -> Reshape
microsoft/onnxscript#2736 · opened on Dec 17, 2025
2
Error with PyTorch exporter: Decorator script does not work on dynamically compiled function Rank
microsoft/onnxscript#2726 · opened on Dec 11, 2025
4
Unintuitive behavior: Returned input aliases generate Identity nodes with generic names instead of preserving Python variable names
microsoft/onnxscript#2714 · opened on Nov 21, 2025
1
[torchlib] Add op test to torch.unique_consecutive
microsoft/onnxscript#2695 · opened on Nov 10, 2025
1
Remove usage of python typing for programmatic annotation of tensor dtype/shapes
microsoft/onnxscript#2666 · opened on Oct 30, 2025
Error exporting BatchNorm2d when PyTorch==2.6.0, onnxscript==0.5.4
microsoft/onnxscript#2651 · opened on Oct 22, 2025
4
Create ORT fusion rules from opset 23 to custom ops
microsoft/onnxscript#2636 · opened on Oct 16, 2025
2
Adding support for integrating arbitrary ModelProto in onnxscript
Feature
microsoft/onnxscript#2616 · opened on Oct 9, 2025
1
Matching variadic lists of tensors?
Feature
microsoft/onnxscript#2581 · opened on Sep 29, 2025 · 0.6.0
12

Navigation Menu

Issues

Issues

Always optimize back-to-back reshapes

Always optimize back-to-back transposes

constant folding does not fold SplitToSequence layer w/o a split input

`@script()` with no explicit opset silently exports functions under a placeholder domain (`this`)

Constant folding uses onnx's plain reference ops, never onnx.reference.ops_optimized

check_model in _framework_apis is a no-op — torch.onnx.export(verify=True) reports success without validating anything

BatchNormalization layer with training_mode=1 results in incorrect ONNX model

Follow up on bincount + index_put

Invalid reshape after FoldConstantsPass in specific {Shape->Gather->Concat->Reshape} pattern

Version converter does not apply applicable conversion

Existing project using v0.5.x fails due to functions in the model

aten::conv3d onnx export bug when bias is missing

CI : Pin github actions to full commit SHA

Constant folding override

Verbose setting for RewritePass for better logging and debugging

Should we turn trace_only default to True in torchlib?

[Optimizer] Fold Shape -> {Slice, Gather} -> Concat -> Reshape

Error with PyTorch exporter: Decorator script does not work on dynamically compiled function Rank

Unintuitive behavior: Returned input aliases generate Identity nodes with generic names instead of preserving Python variable names

[torchlib] Add op test to torch.unique_consecutive

Remove usage of python typing for programmatic annotation of tensor dtype/shapes

Error exporting BatchNorm2d when `PyTorch==2.6.0`, `onnxscript==0.5.4`

Create ORT fusion rules from opset 23 to custom ops

Adding support for integrating arbitrary ModelProto in onnxscript

Matching variadic lists of tensors?