Add PIP installation for loadgen

[gfursin]

As we discussed during the last MLPerf inference meeting, it will be useful to add PIP installation for MLPerf loadgen.

@gaziqbal - let's sync on that since it can simplify our CM automation workflow to run any ONNX model with loadgen.

@arjunsuresh - we can also simplify our CM script to install loadgen via PIP instead of pulling the whole inference repo and building loadgen from there ... Let's discuss how to do it during our next conf-call.

CC @tjablin and @guschmue.

[arjunsuresh]

This is being done here

[arjunsuresh]

This is the pypi release: https://pypi.org/project/mlperf-loadgen-cb

[gfursin]

@arjunsuresh . Hmm, this release looks very outdated (Mar 18, 2022). I would not recommend to use it unless it's regularly updated ...

[arjunsuresh]

Sorry. I shared wrong link. This is the actual work. But I guess the official release is awaiting the code freeze for 3.1

[gfursin]

Cool. Thank you for the info @arjunsuresh . When it's published, we can add the support to our CM scripts! It will simplify dependencies ...

[arjunsuresh]

We do have it now.

pip install mlcommons-loadgen

But this pypi release is not yet managed automatically and so we may not have the latest loadgen here.