This page now serves to manage the files which serve as a source for embeddings to vector database, i.e. source of data for semantic search.

As we know, for doing semantic search over the file, it is required to chunk the file, and generate embeddings based on the chunks, which are then getting inserted into the vector database.

Each tier at Laminar has a limited number of embeddings per month. So it’s best to re-use the embeddings which have been generated before. So in case you have 2 different nodes using embeddings from the same file, re-use the file from “files” page rather than uploading the file again and, hence, generating same embeddings again.

File can be uploaded by going to “files” page and clicking “New file”. When file is selected, it automatically gets uploaded and embeddings are generated. We currently use our default chunking algorithm and model, however, if you need a personalized one, please contact us.

Use case: Semantic Search datasource

We mainly created “files” page, so that you re-use them as a datasource in Semantic Search node.

Semantic Search node can have multiple datasources, which can be added by pressing “Add datasource”.

You can either select a file or upload a file. We strongly suggest to select a file, if it has been uploaded before. This is due to the fact that uploading a file automatically will chunk it and generate embeddings.

About datasets

Note that when you work with datasets, indexing them requires embeddings’ generation too. That is managed in the “datasets” page by pressing “Index” button (Read more)

When you add datapoints to datasets from a file, we only extract datapoints, and we do not generate any embeddings behind the scenes. If you want to generate embeddings for dataset, this is not done automatically, and requires pressing an “Index” button.