How could reuse embeddings generated by openai?

[Tongzhao9417]

I have a quick question: I use the openai embedding as backend. It's successed if I followed the guide. However, It's a little expensive if I request openai embeddings when I modify the hyper-parameter everytime, so I want to save the text embeddings for reuse.

I save the embeddings at a variable embeddings, and create the topic_modeling. But there are some errors: object of type 'numpy.float64' has no len() Here are some codes I wrote:

client = openai.OpenAI(
    base_url="",
    api_key="",
)
embedding_model = OpenAIBackend(client, "text-embedding-ada-002")
embeddings = embedding_model.embed(docs)

topic_model = BERTopic(
    vectorizer_model=vectorizer_model,
    umap_model=umap_model,
    hdbscan_model=hdbscan_model,
)
topics, probs = topic_model.fit_transform(cut_word,embeddings=embeddings) 

[Tongzhao9417]

Solved. I should pass the embedding_model=embedding_model into BERTopic. Here is the full of BERTopic

topic_model = BERTopic(
    embedding_model=embedding_model,
    vectorizer_model=vectorizer_model,
    umap_model=umap_model,
    hdbscan_model=hdbscan_model,
)

Another question is that when I switch the backend model, it's invalid until I restart the jupyter kernal. But its a tiny issues. Thanks for Bertopic!