L3.3 70B Euryale v2.3

L3.3 70B Euryale v2.3 is Sao10K's December 2024 roleplay and creative writing model, fine-tuned on Meta's Llama 3.3 70B Instruct. Unlike the v2.2 release it is trained entirely on top of Llama 3.3 Instruct rather than assembled from an extracted LoRA, which Sao10K positions as the reason it edges out its predecessor. The training mix combines unalignment data, roleplay and creative instruction sets, and the Hesperus-v2 collection, aimed at character consistency and narrative coherence across long exchanges rather than benchmark scores. It is a direct replacement for Euryale v2.2 in existing setups. Sao10K recommends the Llama 3 Instruct prompt format with temperature around 1.1 and min_p near 0.1. It suits interactive fiction, roleplay platforms, and dialogue-heavy applications where persona stability matters more than assistant-style helpfulness.