7 ms·
I saw another repo use weights from llama3 to construct embedding model. I think the use case will be use this small model to search and use bigger model to re-
by tuanmount2 2y ago
I saw another repo use weights from llama3 to construct embedding model. I think the use case will be use this small model to search and use bigger model to re-rank later. So the question will be how this approach compared to BM25
- stephantul 2y agoHey, you are thinking about wordllama. We explicitly compare to wordllama in the results. Feel free to check it out. Your point about BM25 is very interesting. We can run some comparisons! Thanks