Description
Tracker issue for torch.onnx PRs for the 1.13 Release.
- [ONNX] Update
unconvertible_ops#85595 - [ONNX] Create GraphContext and load
g.opmethod to the class #84728 - [ONNX] Create decorator to handle symbolic context #84776
- [ONNX] Annotate all g as GraphContext #85491
- [ONNX] Deprecate setter functions for global variables #85165
- [ONNX] Introduce SARIF #85428
- [ONNX] Migrate SARIF from attr to dataclasses #85651
- [ONNX] Initial version of diagnostics infrastructure. #85107
- [ONNX] Update documentation #85608
- [ONNX] Update behavior for
register_custom_op_symbolic#85636 - [ONNX] Fix reduce node shape inference #85765
- [onnx] Add support for autograd function inlining in ONNX_ATEN_FALLBACK mode #85736
- [ONNX] Exporting the operator prim::type to ONNX opset version 16 is not supported. #77671 torch vision batched_nms
- ONNX tests fail after torchvision update #81121
- ONNX export fails with a
resolve_conjop when a tensor slice is printed. #73619
Likely to be fixed for 1.13
Unable to land due to Meta's internal limitations
Meta will not merge the ONNX bump for 1.13 because it causes issues for them internally that are beyond our control.
Unlikely to be fixed for 1.13
-
[ONNX] Support exporting new dispatch to linalg.vector_norm and linalg.matrix_norm from torch.norm #83661Backlogged, PR will not be in 1.13. - onnx export fails when using torchvision.transforms.CenterCrop #76370
- [ONNX] Error when exporting adaptive_max_pool2d to ONNX #75383
- [ONNX][bug]
nn.Transformercontains unsupported tensor scalar type #85116 - DISABLED test_keypoint_rcnn (__main__.TestONNXRuntime_opset11) #66528
- [ONNX] Enable test_fill script test #79555
- [ONNX] Floor division in ONNX is inconsistent with eager after #78411 #78442
- [ONNX] onnx.export fails on frozen top-level ScriptedModule #86325
cc @BowenBao @abock @thiagocrepaldi @shubhambhokare1 @AllenTiTaiWang @MaanavD
Please feel free to add or modify.