Can't install model_training

[CloseChoice]

Since the commit https://github.com/LAION-AI/Open-Assistant/pull/2206 I can't install model_training anymore. It looks like we try to install ray from a private aws artifactory via trlx.

The error (while just trying to install trlx):

Collecting trlx@ git+https://github.com/CarperAI/[email protected]
  Cloning https://github.com/CarperAI/trlx.git (to revision v0.6.0) to /tmp/pip-install-2_26mpi5/trlx_9eaa446e0cdf486eac6f5fef72af0f93
  Running command git clone --filter=blob:none --quiet https://github.com/CarperAI/trlx.git /tmp/pip-install-2_26mpi5/trlx_9eaa446e0cdf486eac6f5fef72af0f93
  Running command git checkout -q b7db6f9e74c7d8dc719255b27968d2994836957a
  Resolved https://github.com/CarperAI/trlx.git to commit b7db6f9e74c7d8dc719255b27968d2994836957a
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Installing backend dependencies ... done
  Preparing metadata (pyproject.toml) ... done
Collecting ray@ https://ray-ci-artifact-branch-public.s3.amazonaws.com/42bb0357a6fb13e4994789c824f3623f32869ad8/tmp/artifacts/.whl/ray-3.0.0.dev0-cp310-cp310-manylinux2014_x86_64.whl
  ERROR: HTTP error 404 while getting https://ray-ci-artifact-branch-public.s3.amazonaws.com/42bb0357a6fb13e4994789c824f3623f32869ad8/tmp/artifacts/.whl/ray-3.0.0.dev0-cp310-cp310-manylinux2014_x86_64.whl
ERROR: Could not install requirement ray@ https://ray-ci-artifact-branch-public.s3.amazonaws.com/42bb0357a6fb13e4994789c824f3623f32869ad8/tmp/artifacts/.whl/ray-3.0.0.dev0-cp310-cp310-manylinux2014_x86_64.whl from https://ray-ci-artifact-branch-public.s3.amazonaws.com/42bb0357a6fb13e4994789c824f3623f32869ad8/tmp/artifacts/.whl/ray-3.0.0.dev0-cp310-cp310-manylinux2014_x86_64.whl (from trlx@ git+https://github.com/CarperAI/[email protected]) because of HTTP error 404 Client Error: Not Found for url: https://ray-ci-artifact-branch-public.s3.amazonaws.com/42bb0357a6fb13e4994789c824f3623f32869ad8/tmp/artifacts/.whl/ray-3.0.0.dev0-cp310-cp310-manylinux2014_x86_64.whl for URL https://ray-ci-artifact-branch-public.s3.amazonaws.com/42bb0357a6fb13e4994789c824f3623f32869ad8/tmp/artifacts/.whl/ray-3.0.0.dev0-cp310-cp310-manylinux2014_x86_64.whl

cc @sanagno

[sanagno]

Hey thanks, seems ray's location has been changed. Made a PR.

I will get back to you. In the meantime you can install ray manually first.

[dwyatte]

In the meantime you can install ray manually first.

Turns out this doesn't work because of the pinned version of trlx from git. We should be able to pin to https://github.com/CarperAI/trlx/commit/fa991ce244d2f6d1ff63f4348735b7d3e8e40b88 instead. Can someone confirm this version of trlx works?

[andreaskoepf]

Should be fixed by #2984

[CloseChoice]

I still get an installation error and the reason is that we need transformers>=4.28 since we are using llama (see transformers release notes) but trlx just supports transformers versions up to 4.27.1

[olliestanley]

I still get an installation error and the reason is that we need transformers>=4.28 since we are using llama (see transformers release notes) but trlx just supports transformers versions up to 4.27.1

It looks like the underlying issue which caused them to pin transformers<=4.27.1 has already been fixed here. I have left a comment asking if they can remove this restriction.

Update: PR 465 on trlx will fix this soon

[olliestanley]

@CloseChoice trlx has now merged the PR to unpin version