Private hf_cache fix
#663
Merged
Private hf_cache fix
#663
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
squidarth
approved these changes
Sep 15, 2023
| @@ -17,6 +17,6 @@ RUN pip install -r /app/cache_requirements.txt --no-cache-dir && rm -rf /root/.c | |||
| COPY ./cache_warmer.py /cache_warmer.py | |||
| {% for repo, hf_dir in models.items() %} | |||
| {% for file in hf_dir.files %} | |||
| RUN python3 /cache_warmer.py {{file}} {{repo}} {% if hf_dir.revision != None %}{{hf_dir.revision}}{% endif %} | |||
| {{ "RUN --mount=type=secret,id=hf-access-token,dst=/etc/secrets/hf-access-token" if use_hf_secret else "RUN" }} python3 /cache_warmer.py {{file}} {{repo}} {% if hf_dir.revision != None %}{{hf_dir.revision}}{% endif %} | |||
There was a problem hiding this comment.
could you pull hf-access-token into a constant, and then pass it in via templating? We can't share constants w/ Baseten easily, but it still might make this more manageable as a constant
squidarth
reviewed
Sep 15, 2023
| @@ -243,3 +244,20 @@ def test_truss_server_caching_truss(): | |||
| ) | |||
| time.sleep(15) | |||
| assert "Downloading model.safetensors:" not in container.logs() | |||
|
|
|||
|
|
|||
| def test_hf_cache_dockerfile(): | |||
There was a problem hiding this comment.
should this be marked as an integration test?
aspctu
pushed a commit
that referenced
this pull request
Sep 15, 2023
* Always upgrade when pip installing during patch (#659) * Always upgrade when pip installing. * Bump pyproject. * Private `hf_cache` fix (#663) * add fix * bump pyproj * update cache to read secret from correct path * clean up pyproject.toml * add test to check for mount * template out hf token file name * Bump version to 0.7.7 --------- Co-authored-by: Sidharth Shanker <sid.shanker@baseten.co> Co-authored-by: Varun Shenoy <vnshenoy@stanford.edu>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Around a week ago, we noticed that builds around private Hugging Face repos started to break due to not being able to find the access token in the build.
There are two fixes that this PR introduces:
--mountdirective in the dockerfile for caching. This ensures that the cache warmer can pull the secret from an external mount. To support tgi/vllm/etc., propagate theuse_hf_secretvariable in the Dockerfile. This will add the--mountflag when we need a secrets mount.hf_access_tokentohf-access-tokento match with our backend settings.config.yml: passhf_access_tokento the Dockerfile and programatically add it as an env variable if it is set. This is insecure in the context of a build (in a build we read from a secrets file) but is fine for testing.TODO:
--mountis not dropped from the dockerfile