information retrieval

The Tiny Retriever vs OpenAI RAG showdown! Did the fine-tuned underdog beat the champ? See the nDCG results, the tech stack (GPL, SBERT), and why losing isn’t the end.

While my model is training, let’s take a moment to pause and reflect on the process so far, its thorns and roses, and make a few more Bert puns while we’re at it!

What makes a good retriever for RAG? Join me on a journey from classic BM25 to the world of BERT, DistilBERT, fine-tuning tricks like GPL, and why these ‘smaller’ models still punch above their weight.

Why use giant LLMs for RAG? This post kicks off a mission to build a tiny, specialized information retriever using BERT, aiming for better performance on niche/non-English data, faster speeds, and no GPU required!