L31 70B Euryale V2.2

L31 70B Euryale V2.2 is a catalog entry for Sao10K's Euryale v2.2, a creative roleplay model built on Meta's Llama 3.1 with 70 billion parameters, here exposed with an 8K context window. It was fine-tuned over two epochs, with the first focused on multi-turn conversational instruction and the second on creative writing and roleplay data drawn from a mix of human and Claude-generated examples. The training mix includes datasets targeting reasoning and spatial awareness, which support coherence in complex character motivations, narrative branching, and world-building. It also adds more roleplay and creative writing examples than the v2.1 release, improving system prompt adherence. Use it for interactive fiction, roleplay systems, and creative applications where consistent character behavior across long exchanges matters. Sao10K suggests a temperature around 1.2 with min_p near 0.2 and the Llama 3 Instruct format.