mirror of https://github.com/llvm/torch-mlir
25a2c8bd85 | ||
---|---|---|
.. | ||
README.md | ||
builder.py |
README.md
Future Work for Lazy Tensor Core
In the last part of the section Understand The Metrics Report, it is mentioned that after running the metrics report,
If you see
aten::
ops other thannonzero
and_local_scalar_dense
, that usually means a missing lowering in the accelerator plugin.
Looking at the sample output and the sample output produced by running a ResNet18 model and a MaskRCNN model, respectively, on the Lazy Tensor Core using the TorchScript backend, the following operations are needed and not yet supported by the backend:
aten::convolution_overrideable
aten::max_pool2d_with_indices
aten::mean.out
aten::sort
aten::arange.start_out
aten::bitwise_and.Tensor_out
aten::clamp.out
aten::exp.out
aten::index.Tensor
aten::nonzero
aten::rsqrt.out
aten::sigmoid.out
aten::topk.values
aten::upsample_nearest2d.out
Note: This list is incomplete because currently the MaskRCNN example crashes halfway through when run on LTC. The output error can also be found in the MaskRCNN sample output.