Connected Chroma, chroma/sync-workspace fails about vector-admin HOT 5 CLOSED

How was this data embedded? That seems to indicate that it cannot find a document title for the embedding in the metadata. Tools like Llama Index and Langchain should attach the appropriate metadata

from vector-admin.

The data was embedded using Chroma with a local embedder from NodeJS. There is no metadata with the embedding, just document and ids. So I connect from code to the ChromaDB, attach to a collection and insert documents.

Like so:
await collection.add({ ids: ["id1", "id2", "id3"], documents: ["lorem ipsum...", "doc2", "doc3"], })

from vector-admin.

Can confirm that rebuilding the collection with the metadatas: [{"title":"title"}] does work, although it does surprise me that this breaks the import of data, as it could use id instead of title if title is not present?

from vector-admin.

Can confirm that rebuilding the collection with the metadatas: [{"title":"title"}] does work, although it does surprise me that this breaks the import of data, as it could use id instead of title if title is not present?

Absolutely, will mark this as a bug because that's a very specific thing to expect. We will eventually not enforce any specific metadata at all and the table will be customizable to which columns you would want to display at all. Also add/update existing metadata fields as well

from vector-admin.

moved to #40

from vector-admin.