Can't build 2.5.0 with Jetpack 6.1 / CUDA 12.6; documentation is incorrect.

[JC3]

The documentation at https://pytorch.org/TensorRT/getting_started/jetpack.html states (at the bottom):

Please make sure to build torch_tensorrt wheel file from source release/2.5 branch

However, the file toolchains/jp_workspaces/MODULE.bazel.tmpl doesn't exist in this branch, and so the subsequent instructions for generating a MODULE.bazel file are invalid. Also the default MODULE.bazel in this branch does not work for this build. The documentation is not correct, and the provided workflow cannot be used to build TensorRT.

How do I build TensorRT 2.5.0 on a Jetson (ARM64, Jetpack 6.1 CUDA 12.6 PyTorch 2.5.0)?

Can't use 2.6.0 because there's no Jetson build of PyTorch 2.6.0, only 2.5.0.

[lanluo-nvidia]

@JC3
Here is the jetPack6.1 build PR: https://github.com/pytorch/TensorRT/pull/3211/files
Which has been merged to main only, what you can do is cherry pick the commits in this PR to release/2.5 branch.

[lanluo-nvidia]

meanwhile I will also cherry pick this into release/2.5 branch

[JC3]

@lanluo-nvidia Thanks for this. I was also able to get it to build after hacking on setup.py and MODULE.bazel for a bit, was about to report back here.

But that PR is very useful, thank you.

[lanluo-nvidia]

@JC3 just to let you know the PR has been cherry picked into release/2.5
#3285

[lanluo-nvidia]

close it.