fix(local): read the size in a Featherless-shaped listing row's id - #2029
Conversation
surfsense_local/backend/modules/llm/profile/fingerprint.py: from_remote() now checks the id for a stated size before falling back to a vendor, so a row with no hugging_face_id classifies on size when its id states one. An empty owned_by and no slash in the id now leaves vendor unset instead of an empty string. surfsense_local/backend/tests/unit/llm/profile/test_fingerprint.py: add a Featherless-shaped row and a slashless id with no size or vendor. docs/architecture/local-models/selection.md: update the Fingerprint Remote bullet and delete the closed Known gaps line.
|
@jamalkamaladdin is attempting to deploy a commit to the Rohan Verma's projects Team on Vercel. A member of the Team first needs to authorize it. |
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
What
modules/llm/profile/fingerprint.py: from_remote reads a stated size out of a listing row's id when it has no hugging_face_id.
An empty owned_by and no slash in an id leaves vendor unset instead of an empty string.
Why
Fixes #1988
How to test
High-level PR Summary
This PR fixes how the system extracts model size and vendor information from API listing rows that lack a
hugging_face_idfield. Previously, models from providers like Featherless (which don't include HuggingFace IDs) were misclassified as frontier models regardless of their actual size. The fix enables reading size information directly from the model ID and properly handles edge cases like empty vendor names, ensuring models likeqwen3-4bare correctly identified as 4B parameter models instead of being treated as frontier-tier.⏱️ Estimated Review Time: 5-15 minutes
💡 Review Order Suggestion
surfsense_local/backend/tests/unit/llm/profile/test_fingerprint.pysurfsense_local/backend/modules/llm/profile/fingerprint.pydocs/architecture/local-models/selection.md