| Updates to Quantization documentation |
30372 |
Yes |
Documentation |
Complete |
| Specify ordering on singular values and eigenvalues output from torch… |
30575 |
Yes |
Documentation |
Complete |
| Fix BC for quantized linear |
30629 |
Yes |
BC Fix |
Complete |
| torch.where changes made on 1.3.1 but not on master |
30729 |
Yes |
Serious bug if not fixed |
Complete |
| Update persons_of_interest.rst |
30648 |
Yes |
Documentation |
Complete |
| Support exporting aten::copy_ and aten::index_put to ONNX opset 11 |
30939 |
No |
Not necessary for torchvision in this release |
Skipped |
Fix reading __cuda_array_interface__ without strides |
30901 |
No |
Superseded by 31450
|
Skipped |
| Fix exception message in Java Tensor |
30776 |
Yes |
Trivial Fix |
Complete |
| turn off profiling graph exec |
30750 |
Yes |
Feature flag not ready for 1.4 |
Complete |
| Dump operator names of a script module |
30747 |
Yes |
Prereq for custom mobile build script tutorials |
Complete |
| Restructure docs organization and naming |
30841 |
No |
The PR doesn't look correct and it's not critical that doc fixes are merged by RC time |
Skipped |
| Fix copy kernel regression for v1.4.0 |
31322 |
Yes |
Regression |
Complete |
[jit] Make nn.Transformer TorchScript compatible |
31390 |
No |
Feature, not in that long |
Skipped |
| [ONNX][v1.4.0] Cherry-pick Add ONNX Export Support to floor_divide |
31397 |
No |
floor_divide doesn't exist in 1.4. |
Skipped |
| Fix coverage and hypothesis conflict |
31429 |
Yes |
caffe2 tests fail otherwise |
Complete |
| Fix unflatten when dim is a negative integer |
31432 |
Yes |
Segfault (only on Named Tensors though) |
Complete |
Fix reading __cuda_array_interface__ inferred strides |
31450 |
Yes |
Bug Fix |
Complete |
| [v1.4.0] Support optional float parameters (float?, optional)
|
31530 |
Yes |
Necessary for MSFT ONNX Patch |
Complete |
| [v1.4.0] Disable flaky TestMomentumSGD.test_fp16momentum_sgd |
31637 |
Yes |
Flaky test |
Complete |
| [v1.4.0] Include two caffe2 ops in v1.4.0 |
31716 |
Yes |
Yes |
Caffe2-only |
| Adding elementwise kernel also operating on index |
31850 |
No |
Introduces another bug (integral mismatch between CPU/CUDA) |
Skipped |
| ONXX interpolate fixes: 31170 and 31526
|
31878 |
No |
Too Complicated |
Skipped |
| [1.4.0] Fix nvcc math functions for MSVC 2019 |
31816 |
Yes |
Correctness bug if built with VS2019 |
Complete |
| Restore CUDA half linspace+logspace and add coverage tests |
31959 |
Yes |
Fixes 30404
|
Complete |
| [v1.4.0] Minimal changes in interpolate to support Keypointrcnn |
31890 |
No |
Will support ONNX export for torchvision |
Superseded by 32010
|
| [v1.4.0] Minimal changes in interpolate to support Keypointrcnn |
32010 |
Yes |
Will support ONNX export for torchvision |
Complete |