There are several key factors to consider when actually deploying Zerank-1:
-
License Restrictions: The main model (zeroentropy/zerank-1) is licensed under a non-commercial license, for commercial use you need to contact ZeroEntropy for a license; and the open source free version (zeroentropy/zerank-1-small) is licensed under the Apache 2.0 license and is available for commercial use.
-
hardware requirement: GPUs are recommended for optimal performance, especially when processing large volumes of documents; CPUs can run but are slower when data volumes are small or latency requirements are low.
-
Explanation of scoresThe relevance score output from the model is a relative value rather than a fixed range, and in practice it is sufficient to compare the scores of the same batch of documents.
-
performance optimization: As a two-stage sorter, usually only the first 100-1000 candidate documents need to be processed, and feeding the entire document library directly into the model computation should be avoided.
In addition, Zerank-1 is more suitable as a complementary component to the existing search system rather than a replacement, and works best in conjunction with fast vector search or keyword search. In the system design, it is necessary to weigh the precision and latency, and rationally configure the ratio of the first stage of retrieval and the second stage of reordering according to the specific business requirements.
This answer comes from the articleZerank-1: A reordering model for improving the precision of search resultsThe
