Medium

Daniel Tunkelang

6.8K followers

All

Daniel Tunkelang

clapped

Jun 27

We deleted fifty years of information retrieval and called it RAG

Chunking, embeddings, hybrid search, reranking, eval. Every one already had a name in information retrieval. We didn’t lose it, we just…

Generative AI

Alvis Ng

Jun 26

We deleted fifty years of information retrieval and called it RAG

Daniel Tunkelang

responded

Apr 20

The products are indexed using the base MiniLM model. The model that's being trained is for queries, and it's a fine-tuned version of the same model used for the products. The only operations at query time are inference of the query on the fine-tuned model and KNN retrieval from the product index. Feel free to check…

Prithivi Da

The precomputed index during training time and inference time are vectorised by which model ? Are you reindexing post training ?

Distilling Retrieval Pipelines to a Single Embedding Model

You're all caught up!