Diego
Changes: - [x] Skip `tests/test_ms_deform_attn.py` when `torch.cuda.is_available()` yields false. - [x] Move `cuda()` calls from global scope to the local scope of tests.
Currently detrex documentation is built from commit df3e6471 (Mar 20, 2023), and there are plenty of fixes from that point onwards that haven't been deployed, to name a few: -...
When running EVA on COCO as per the [docs](https://github.com/baaivision/EVA/tree/master/EVA-02/det#coco-3), Pillow throws an error because the copy of detectron2 uses the deprecated PIL.Image.LINEAR. This issue has been handled in the original...
🐛 [Bug] Importing `torchao` first breaks `torch_tensorrt.dynamo.compile` during `run_decompositions`
## Bug Description Importing `torchao` before importing `torch_tensorrt` causes `F.interpolate` to fail during `run_decompositions` with: `AssertionError: Expected aten.upsample_nearest2d.default to have CompositeImplicitAutograd kernel` ## To Reproduce ```py import torchao import torch_tensorrt...
I'm not sure if this is intended, but it seemed odd to me that there is logic inside both fast and global partitioners (see snippets 1 and 2) with `require_full_compilation`...
## Tasks - [ ] `torch_tensorrt.dynamo.conversion.impl.slice.ops.repeat` - Reference: https://github.com/pytorch/TensorRT/blob/fa02fd3e3a85a6042d11a00cd386f6b69c1d6c4b/py/torch_tensorrt/dynamo/conversion/impl/slice/ops.py#L219-L324 - [ ] `torch_tensorrt.dynamo.conversion.aten_ops_converters.aten_ops_repeat` https://github.com/pytorch/TensorRT/blob/fa02fd3e3a85a6042d11a00cd386f6b69c1d6c4b/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py#L1152-L1172 - [ ] Unit test, [example: expand](https://github.com/pytorch/TensorRT/blob/main/tests/py/dynamo/conversion/test_expand_aten.py)