-
Notifications
You must be signed in to change notification settings - Fork 48
Add huggingface/pytorch/tei/docker/1.7.1
#153
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
TEI containers don't have Python, neither PyTorch in consequence, those are Rust only, so those extra fields are not required and could potentially be misleading Also the field `cuda_version` has been removed from one of the CPU variants, as CPU images don't have CUDA installed
- Assume `python_version` and `pytorch_version` can be `None` i.e., Optional, since those are not required for Text Embeddings Inference (TEI) - Fix typing accordingly and run `ruff` to ensure code quality
LOG.info(f"Going to test built image for config: {config}.") | ||
test_role_arn = os.getenv(EnvironmentVariable.TEST_ROLE_ARN.name) | ||
test_session = aws.get_session_for_role(test_role_arn) | ||
test_session = aws.get_session_for_role(test_role_arn) # type: ignore |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why these ignores?
environ = os.environ.copy() | ||
environ.update( | ||
{ | ||
{ # type: ignore |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
same comment here
@alvarobartt curious why are we adding the |
Hey @malav-shastri so I was adding those as the type checkers were complaining which can sometimes be tedious / annoying, and that's not because the types are incorrect, but rather than the types are arbitrary e.g. the same variable can either be |
Also @malav-shastri we need to take into consideration the note I included in #153 (comment), since that will affect the image URIs from now on as those won't come with the misleading tags for Python and PyTorch (as per described in the PR description). So flagging it here just in case, and also cc'ing both @arjkesh and @fgbelidji for visibility! |
Hello @malav-shastri, can you let me know when the TEI 1.7.1 images are available in all the public registries so I do a PR to the Python Sagemaker SDK? |
Description of changes:
This PR adds the Text Embeddings Inference (TEI) 1.7.1 recently released, that comes with some patches that were blocking some AWS SageMaker users as per e.g. huggingface/text-embeddings-inference#596 (comment).
Find the release notes at https://github.com/huggingface/text-embeddings-inference/releases/tag/v1.7.1
Important
This PR also updates the release process for Text Embeddings Inference (TEI) images, since those come with both Python and PyTorch version specifiers, which are misleading and not correct, since the TEI images are Rust-only. In consequence the
release_utils.py
script has been updated accordingly to remove those values from the TEI URIs, as well as from thereleases.json
file and made bothpython_version
andpytorch_version
configuration values optional as in thecuda_version
(which was already optional).This being said, the changes in this PR may affect other repositories pulling the former URIs, which again are incorrect, so happy to help fix it everywhere when needed to make sure that containers released from now on follow the current URI formatting proposed in this PR, as well as upstreaming the fix to the AWS SageMaker SDK et al.
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.