v1.0 drops embeddings_util.py breaking semantic text search

[mrbullwinkle]

Describe the bug

The previous version of the OpenAI Python library contained embeddings_utils.py which provided functions like cosine_similarity which are used for semantic text search with embeddings. Without this functionality existing code including OpenAI's cookbook example: https://cookbook.openai.com/examples/semantic_text_search_using_embeddings will fail due to this dependency.

Are there plans to add this support back-in or should we just create our own cosine_similarity function based on the one that was present in embeddings_utils:

def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

To Reproduce

Cookbook example cannot be converted to use v1.0 without removing the dependency on embeddings_utils.py https://cookbook.openai.com/examples/semantic_text_search_using_embeddings

Code snippets

from openai.embeddings_utils import get_embedding, cosine_similarity

# search through the reviews for a specific product
def search_reviews(df, product_description, n=3, pprint=True):
    product_embedding = get_embedding(
        product_description,
        engine="text-embedding-ada-002"
    )
    df["similarity"] = df.embedding.apply(lambda x: cosine_similarity(x, product_embedding))

    results = (
        df.sort_values("similarity", ascending=False)
        .head(n)
        .combined.str.replace("Title: ", "")
        .str.replace("; Content:", ": ")
    )
    if pprint:
        for r in results:
            print(r[:200])
            print()
    return results


results = search_reviews(df, "delicious beans", n=3)

OS

Windows

Python version

Python v3.10.11

Library version

openai-python==1.0.0rc2

[RobertCraigie]

Thanks for calling this out @mrbullwinkle, we're working on updating the cookbook repository to include the functions provided in embeddings_utils.py directly so that you can copy them into your own project.

This is a better approach than the current embeddings_utils as you can just include the dependencies for the function you want whereas with the current approach you'll have to install dependencies you'll never use.

[mrbullwinkle]

@RobertCraigie, makes sense. Thank you for the super fast response!

[CristianPQ]

@mrbullwinkle
In the meantime, I hope this piece of code can help you.

source: https://learn.microsoft.com/en-us/azure/ai-services/openai/tutorials/embeddings?tabs=python-new%2Ccommand-line

def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

def get_embedding(text, model="text-embedding-ada-002"): # model = "deployment_name"
    return client.embeddings.create(input = [text], model=model).data[0].embedding

def search_docs(df, user_query, top_n=4, to_print=True):
    embedding = get_embedding(
        user_query,
        model="text-embedding-ada-002" # model should be set to the deployment name you chose when you deployed the text-embedding-ada-002 (Version 2) model
    )
    df["similarities"] = df.ada_v2.apply(lambda x: cosine_similarity(x, embedding))

    res = (
        df.sort_values("similarities", ascending=False)
        .head(top_n)
    )
    if to_print:
        display(res)
    return res


res = search_docs(df_bills, "Can I get information on cable company tax revenue?", top_n=4)

[mrbullwinkle]

@CristianPQ you beat me to adding a ref to that code sample to this issue after I added it earlier today. I am the author of that code sample/article. ☺️

[patham9]

Very unprofessional that such functions just get removed. It is a moving API one cannot rely on, very annoying.

[bhaniraj]

@RobertCraigie , is the support added in cookbook repo for the functions provided in embeddings_utils.py ?

[rattrayalex]

@logankilpatrick , can you help here?

[julianlevy]

Hi folks - any updates on the cookbook yet?

[jawwadturabi]

Still can't see the updates on the cookbook

[logankilpatrick]

Hey folks, we migrated these over to the cookbook's own utils folder ~3 months ago: https://github.com/openai/openai-cookbook/blob/main/examples/utils/embeddings_utils.py, if you find any notebooks that are out of sync and not using the built-in utils, please open an issue on the cookbook repo.

[the-rich-piana]

Why the frick would you guys remove this. This API is god awful.

[davidgilbertson]

FWIW, if you're using embeddings from the OpenAI API, you can get cosine similarity with just a @ b, since they're normalized to 1 anyway (assuming a and b are NumPy arrays or tensors or similar, not plain lists)

[davidgilbertson]

Also BTW there is still a reference to embeddings_utils in the actual user guide in the docs here: https://platform.openai.com/docs/guides/embeddings/use-cases