Sparse and lexical vectors
#1
by
PeppePasti
- opened
Given that this model is based on https://huggingface.co/BAAI/bge-m3, it still has the multi-granularity property? Or it has been trained only on the dense vectors part?
In this case here the emphasis is more about optimizing the dense vectors to improve their semantic accuracy for the specific (legal) task.
That said, the relationship between multi-granularity and dense vectors in this model is complementary rather than mutually exclusive: one should expect in this work.
Obiactum
changed discussion status to
closed