Repository navigation
docs: attribute the vendored chat templates and container profiles - #644
Merged
Merged
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
Collaborator
Author
|
Review: the paths are correct, and so is the moby/profiles Apache-2.0 attribution with its upstream pin. Three things before merge:
|
NOTICE now names the third-party files Flynn ships beside the Chroma lexers: the gpt-oss and Gemma 4 chat templates, copied unchanged from the files OpenAI and Google publish on Hugging Face, and the seccomp profile and AppArmor template derived from moby/profiles. All are Apache-2.0. The template variables record the upstream revision each file was taken from, and a test holds each template to the SHA-256 of that upstream file, so the "copied unchanged" statement fails CI rather than going stale if a template is edited in place. The watch package comment describes the ai! and ai? marker syntax directly. Signed-off-by: Ion Alpha <[email protected]>
The Gemma 4 model card (google/gemma-4-12B-it) declares apache-2.0 and links the Gemma 4 licence page, which is the unmodified Apache 2.0 text. The Gemma Terms of Use scope lists only earlier models. NOTICE now says so, so the template and the weights are under the same terms. The gemma4 template comment named the -qat-q4_0-gguf repository as well as the 12B-it repository. The file is byte for byte the chat_template.jinja at the recorded 12B-it revision (sha256 verified), so the comment names that one source. Also wrap two over-long comment lines. Signed-off-by: Ion Alpha <[email protected]>
ion-alpha-dev
force-pushed
the
docs/notice-third-party-files
branch
from
November 1, 2026 09:01
0585e65 to
fd11f39
Compare
This file contains hidden or 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
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
What
NOTICEnow attributes the third-party files Flynn ships: the gpt-oss and Gemma 4 chat templates, and the seccomp and AppArmor profiles derived from moby/profiles. The template variables record the upstream revision each was taken from, and a new test pins each template to its publisher's SHA-256. Thewatchpackage comment now describes theai!/ai?marker syntax directly.Why
All four files are Apache-2.0, which asks redistributors to keep the attribution with the work. Until now
NOTICElisted only the Chroma lexers. The digest test keeps the statement "copied unchanged" true: an in-place edit to a template fails CI instead of quietly serving a prompt format that differs from the one the model was trained on.How to verify
go test ./internal/inference/launch -run TestEmbeddedTemplatesAreThePublishersOwn.jinjafile and the test fails.chat_template.jinjainopenai/gpt-oss-20b(revision6cee5e8) andgoogle/gemma-4-12B-it(revision43c9d01). Both repos are taggedapache-2.0on Hugging Face.Notes for reviewers
NOTICEand a test change.dev/container-profiles.NOTICEpoints there instead of repeating it.