Thanks, requesting for details on instructions
#2
by
gsaivinay
- opened
E5 models are not instruction tuned.
For retrieval tasks, we prepend "query: " and "passage: " to the query and corpus passages, respectively.
For all other tasks, we prepend "query: " to the inputs.
As said in the model card, all the results can be reproduced using code at https://github.com/microsoft/unilm/tree/master/e5