3 ms·
What do you mean by fully analyzed? It’s the LLM that does the embedding.
by wodenokoto 3y ago
What do you mean by fully analyzed? It’s the LLM that does the embedding.
- visarga 3y agoOh the embedding LLMs are usually lightweight BERT models with few layers and <<1B weights, while LLMs are easily 10-100x larger. The idea is to ingest the text in a LLM to extract the facets you are going to search and add those extra tokens to the original text. Then you do regular RAG.
- sroussey 3y agoHow?
- wodenokoto 3y agoYou want to enrich the base text before embedding? With what kind of prompt? It’s a pretty simple thing to add to a pipeline. Have you tried?