This blog post on the effect of adding "structured generation" on the performances and consistency of LLMs was a good read Thanks a lot for that @willkurt @remilouf and @clefourrier

→ View original post on X — @thom_wolf,