How to load pre-generated embedding files?

[rustam-ashurov-mcx]

Hey mates, have a hard time with SK + KM (mostly because of lack of exp of working with both yet) 😅

My aim is to generate embeddings in advance as files, and on service startup upload them in in-memory store without calling AI provider again and again on each restart.

So I generated embeddings via OpenAI client (I was unable to find whether I can generate them via SK/KM and store as files), and now they are stored as *.json files with such (example) content:

"{
"data": [
{
"embedding": [
0.006308248266577721,
....lot of numbers here...
],
"index": 0,
"object": "embedding"
}
],
"model": "text-embedding-ada-002",
"object": "list",
"usage": {
"prompt_tokens": 305,
"total_tokens": 305
}
}"

What I can not understand is how can this data be loaded into KM?

[dluc]

hi @rustam-ashurov-mcx, what you're describing is very similar to a cache, which would require storing pairs of text and vectors, allowing to search by text. You could implement a cache dependency and inject into embedding generators, using a lookup table. Or you could develop a custom embedding generator, that never talks to OpenAI, and always loads embeddings from storage, taking care of all the cases, e.g. distributed storage, adding new embeddings, etc.