polish-reranker-large-ranknet
sdadas/polish-reranker-large-ranknet
A popular open reranker model, with 31.8K downloads a month. gigarouter benchmarks and hosts it as an OpenAI-compatible API.
about this model
This is a Polish text ranking model trained using RankNet pairwise loss on a dataset of 1.4 million queries and 10 million documents. Training data includes the Polish MS MARCO split (800k queries), the ELI5 dataset translated to Polish (500k queries), and a collection of Polish medical question-answer pairs (100k queries). The student model is Polish RoBERTa, distilled from a large multilingual MT5-XXL teacher.
Key Strengths
RankNet loss evaluates relative document order per query rather than scoring each document independently, improving ranking quality. Despite being 30 times smaller and 33 times faster, the model outperforms its teacher on the Polish Information Retrieval Benchmark.
Benchmark Results
| Benchmark | Category | Metric | Score |
|---|---|---|---|
| Polish Information Retrieval Benchmark (PIRB) | Rerankers | NDCG@10 | 62.65 |
Full results are available on the PIRB Leaderboard.
We're benchmarking and onboarding polish-reranker-large-ranknet as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.