Jina AI just released a multilingual reranker model for RAG and retrieval. It's quite efficient, and performs well for English and beyond. Sadly, it has a CC-BY-NC-4.0 license. Details: - Competitive on both English (BEIR, CodeSearchNet) and multilingual benchmarks (MKWA, MLDR, AirBench) - Open weights: can be ran via Sentence Transformers, Transformers, or Jina's API - 278M parameters with Flash Attention 2 to speed up inference: notably faster than alternatives - Also trained on SQL & code data; seems to work well for function calling reranking for RAG + code - Sadly, a non-commercial license 🏛 Check out the model on Hugging Face: https://lnkd.in/e8rrnw9d Congratulations to the Jina team for this release!
Do you data what are their latency numbers? ( “notably faster than alternatives “) Efficient cross encoders would be helping in many search applications esp as they hit those numbers in relevance reported in the release. No commercial license. Very sad. Most search applications are commercial
Founder & CTO @ Nomic
2wWhy is it bad that someone wants to make money off a open model by placing a non-commercial license on it, don't you want to keep them incentived to release more open models?