[backends] List models function call replacement

[gs-olive]

• Replace _list_model_paths call with list_models

[xuzhao9]

Thanks for fixing this!

[xuzhao9]

BTW, we tested the running the userbenchmark at: https://github.com/pytorch/benchmark/actions/runs/5833882739

Unfortunately, the workflow failed because we currently don't have tensorrt or torch_trt installed in our nightly docker.
Can you please help fix our docker file? https://github.com/pytorch/benchmark/blob/main/docker/torchbench-nightly.dockerfile

[facebook-github-bot]

@xuzhao9 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

@xuzhao9 merged this pull request in 030fdbd.

[gs-olive]

Hi @xuzhao9 - thanks for the reference! I looked into the Dockerfile and there's a few considerations we'll need to address on the Torch-TRT side to ensure the testing will work. I have a few questions to address this:

1. At what time is the nightly CI run?
2. What Python and CUDA versions does the Docker container use?
3. Does the Docker container use the latest-available Torch nightly?

This will help me to understand the requirements to provide a nightly build of Torch-TRT which will be compatible with the versions in the container.

[xuzhao9]

@gs-olive
See my answers below:

1. The nightly CI will run everyday at 4PM UTC (https://github.com/pytorch/benchmark/blob/main/.github/workflows/userbenchmark-a100.yml#L4) after the latest nightly docker is built.
2. The default Python and CUDA versions are defined at https://github.com/pytorch/benchmark/blob/main/utils/cuda_utils.py#L11 and https://github.com/pytorch/benchmark/blob/main/utils/python_utils.py#L4, respectively. It is easy to upgrade them and follow the recommendation of the latest pytorch nightly build.
3. Yes, the Docker container uses the latest-available Torch nightly release, we will update the Docker everyday at https://github.com/pytorch/benchmark/actions/workflows/build-nightly-docker.yml

[gs-olive]

Thank you for the detailed answers, this is very helpful. We are working on publishing nightly releases of Torch-TRT via Github Actions and I will update the Dockerfile as soon as those are ready.