Where is the distances_from_embeddings helper function in the updated OpenAI Python library?

Below is a brief explanation and a sample solution.

What’s Happened?

The helper function you mentioned was removed from the library. In the updated releases, similar functionality is expected to be implemented by you directly. This change is part of a broader simplification, especially as OpenAI embeddings are normalized.

How to Replace It

Since OpenAI embeddings are normalized to unit length, you can compute cosine similarity with a simple dot product. For example:

PYTHONCopied
import numpy as np

def cosine_similarity(embedding1, embedding2):
    # Since both embeddings are normalized to 1, the dot product is the cosine similarity.
    return np.dot(embedding1, embedding2)

# Example embeddings (assume these are from your API call)
embedding_a = [0.1, 0.2, 0.3, 0.4]
embedding_b = [0.2, 0.1, 0.4, 0.3]

similarity = cosine_similarity(embedding_a, embedding_b)
print("Cosine Similarity:", similarity)

Why a Dot Product Works

Because the embeddings are normalized (i.e., they have a length of 1), the cosine similarity reduces to the dot product. This means you no longer need a dedicated helper function—computing the dot product with a library like NumPy is both efficient and straightforward.

Conclusion

By computing the dot product yourself, you effectively replace the removed helper function. This approach gives you flexibility and helps you stay current with the updated API.

If you have any more questions about using embeddings or other OpenAI API topics, feel free to ask!