Chao Zhang

Above should be reproducible on NGC container `pytorch:22.06-py3`

Currently working on this implementation, but it's a bit tricky.

Adding to this, `tuple` arguments have the same behavior _iff_ combined with `...` or `:`. ie. ``` t = torch.rand(3, 4, 5) t[(1, 2)] == t[1][2] # This should be...

Alright I think I have an algorithm that should work in the general case, I'll document it on the PR...

https://github.com/NVIDIA-AI-IOT/torch2trt/pull/783

Note that this was tested on v0.3.x; the unit tests here fail on v0.4.0. See #769.

Running `python -m torch2trt.test --name getitem`: ``` NUM_TESTS: 39 NUM_SUCCESSFUL_CONVERSION: 39 NUM_FAILED_CONVERSION: 0 NUM_ABOVE_TOLERANCE: 0 NUM_pSNR_TOLERANCE: 0 ```

``` NUM_TESTS: 38 NUM_SUCCESSFUL_CONVERSION: 38 NUM_FAILED_CONVERSION: 0 NUM_ABOVE_TOLERANCE: 0 NUM_pSNR_TOLERANCE: 0 ```

Yeah, when fallback occurs due to unsupported operators, this already prints out the unsupported operator as you mentioned with the file and line number (eg. https://github.com/NVIDIA/Torch-TensorRT/issues/803). For this request, I...

Hmm yeah that's fair; I'm in agreement with what's stated here, especially with regards to not sending users off on a wild goose chase. That said, would you say that...