InRanker 3B (3B parameters) InRanker is a version of monoT5 self distilled from monoT5 3B with increased effectiveness on out of domain scenarios. Our key insight were to use language models and rerankers to generate as much as possible synthetic "in domain" training data, i.e., data that closely resembles the data that will be seen at retrieval time. The pipeline used for training consists of two distillation phases that do not require additional user queries or manual annotations: (1) training on existing supervised soft teacher labels, and (2) training on teacher soft labels for synthetic queries generated using a large language model. The paper with further details can be found here. The code and library are available at https://github.com/unicamp dl/InRanker Usage The library was tested using python 3.10 and is installed with: The code for inference is: How to Cite
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy