Thank you for the Mizan rerankers — excited to try V3-Turbo!
A thank-you to the developer of the Mizan reranker models, with a few questions about the V3-Turbo training methodology.
Assalamu alaykum wa rahmatullahi wa barakatuh, Ali.
I wanted to thank you for all the effort you've put into developing the Mizan reranker models. I've been using Mizan for some time, and it has been genuinely useful in my projects. I'm really excited to try the new V3, especially V3-Turbo!
I'm currently developing a RAG system over around 40,000 fatwas from trusted Salafi scholars, expanding my original Ibn Baz project into a multi-scholar system. Here's my GitHub repository: https://github.com/AbderrahimZineddineEsi/BinBaz-RAG
Your work is particularly relevant to what I'm building, and I'm excited to evaluate Mizan-Rerank-v3-Turbo in my retrieval pipeline.
JazakAllahu khayran for sharing your work with the community. May Allah reward you for the benefit it brings to Arabic NLP and Islamic knowledge.
If you have time, I'd also love to learn more about how you built the model, especially the training datasets, hard-negative generation, and distillation process.
BarakAllahu feek.