Follow up PR for Audio End to End testing #390
Conversation
|
Codecov ReportAll modified and coverable lines are covered by tests ✅
❗ Your organization needs to install the Codecov GitHub app to enable full functionality. Additional details and impacted files@@ Coverage Diff @@
## main #390 +/- ##
=======================================
Coverage 78.99% 78.99%
=======================================
Files 37 37
Lines 2804 2804
=======================================
Hits 2215 2215
Misses 589 589 ☔ View full report in Codecov by Sentry. |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
PR Summary
This PR enhances the audio embeddings functionality in the infinity_emb library, addressing issue #378 by improving support for mixed input types and expanding test coverage.
- Modified
AudioEmbeddingInputinpymodels.pyto accept both text and URL inputs - Extended
_embeddings_audiofunction ininfinity_server.pyto process mixed input types (audio URLs and text) - Added new tests in
test_torch_audio.pyfor text-only input, mixed text/URL input, and embedding comparisons - Implemented cosine similarity function for comparing audio embeddings
- Removed skip decorator from text-only test, improving overall test coverage
3 file(s) reviewed, 3 comment(s)
Edit PR Review Bot Settings
| assert cosine_similarity( | ||
| embeddings_audio_beep, embeddings_text_beep | ||
| ) > cosine_similarity(embeddings_audio_beep, embeddings_text_fish) | ||
| assert cosine_similarity( | ||
| embeddings_audio_beep, embeddings_text_beep | ||
| ) > cosine_similarity(embeddings_audio_beep, embeddings_text_horse) |
There was a problem hiding this comment.
style: Add a small tolerance to comparisons to account for floating-point precision
There was a problem hiding this comment.
Just a greater than is fine here I guess
|
@wirthual Wait, I am confused - the “/embeddings” endpoint already works for audio! |
Isnt the |
#378 This PR adds support for handing text and urls for the audio embeddings endpoint.
Extended the test to test text only, mixed text and urls and also check a meta test for the embeddings.
Implemented for the audio endpoint, if this scheme is acceptable happy to port it to the vision endpoint.
Cheers