use embeddings that match OpenAI completion vector size: 1536 (medium) instead of 3072 (large)

[austinworks]

true with floats

and also base64 representation

[carl]

rounding to 3 digits is still not the answer:

[paulz]

we currently use large model for embedding, should try medium:

See:

> outside_tolerance_count = np.sum(np.abs(diff_val) >= 0.1)

[nitpick] Avoid using the magic number 3072 directly; consider defining a constant or adding a comment explaining its significance to improve code clarity.
assert len(embedding_object["embedding"]) == 3072
examples/team_recommender/tests/example_1_text_response/test_compute_cosine_similarity.py:97

[nitpick] Document the rationale behind updating the tolerance thresholds and the expected count (from 900 to 952) to clarify the intended behavior of the test.
outside_tolerance_count = np.sum(np.abs(diff_val) >= 0.1)