L3 70B Euryale V2.1

L3 70B Euryale V2.1 is Sao10K's creative roleplay model built on Meta's Llama 3 with 70 billion parameters and an 8K context window, aimed at character-driven narratives and dialogue. It was produced through LoRA fine-tuning on 8x H100 SXM hardware using the same dataset as Stheno v3.2, with a focus on creative writing and roleplay. The model shows improved prompt adherence and spatial awareness over smaller alternatives, adapting to custom formatting and character definitions while keeping creative output consistent. Sao10K's recommended settings are temperature 1.17, min_p 0.075, and repetition penalty 1.10. This is the earlier release that precedes v2.2, which moves to Llama 3.1 and a longer context. Reach for v2.1 when an 8K context and a Llama 3 base fit the workload, across roleplay platforms, creative writing assistants, and dialogue-heavy applications.