Thibaut Goetghebuer-Planchon comments

Results 128 comments of


                                            Thibaut Goetghebuer-Planchon

Add support for upsample_nearest2d op in the Arm backend

Thanks for the review. I rebased the change and fixed the conflicts.

Add support for upsample_nearest2d op in the Arm backend

Thanks, the last change should fix this.

Add support for upsample_nearest2d op in the Arm backend

Thanks, I rebased and fixed the issue.

Add support for upsample_nearest2d op in the Arm backend

@digantdesai Thanks, done

Add support for upsample_nearest2d op in the Arm backend

The failure doesn't seem to be related to this PR as it's the new `test_var.py` tests introduced during the rebase from upstream that are failing. I'll check internally if these...

Add support for upsample_nearest2d op in the Arm backend

The unit test failure is not due to the code of this PR, we confirmed it internally. We should check to raise another PR to fix the `torch.var` test failure...

Add support for upsample_nearest2d op in the Arm backend

With https://github.com/pytorch/executorch/commit/995c2bf0392ac3843e5e255ac5d06538458d0b2c fixing the test_var accuracy issue, the tests are now passing.

The memory usage increases sharply

Do you use the identity hash function? That would create a lot of collisions with the default map, which uses a power of two for the size of the table,...