Fix load index

[buhe]

Hi, @ZachNagengast

The loadIndex method has already loaded the embedding, and there is no need to call addItem. The original method is very slow in some cases. FIY https://github.com/buhe/langchain-swift/blob/main/Sources/LangChain/vectorstores/SimilaritySearchKit.swift #L35 with openaiembedding

BR
buhe

[ZachNagengast]

Nice! Do you think we can still add these to the index object after calling it? This function kind of does two things, one is loading the embeddings from the store and one is adding them to the parent object for searches. Definitely support splitting these apart, but curious how you think it could be handled if someone wants to load the items and then run a search of them?

[buhe]

I I've added it and tested it :)
https://github.com/ZachNagengast/similarity-search-kit/pull/36/files
Please look carefully, the results of assignment are different. : )

[ZachNagengast]

I think addItems is necessary to make sure the index has access to all the embeddings from the vector store. When using this function it should be possible to do this:

        let _ = try! similarityIndex.loadIndex(name: "TestIndex")
        let results = similarityIndex.search("search query for TestIndex items")

[buhe]

The modified code works, I tested it, after assigning it to indexItems, addItem is not needed.

[ZachNagengast]

Ah, I understand now, great!

[buhe]

👍

[ZachNagengast]

Looks good, nice find and thanks for submitting, this should help speed things up 🙌