Sao10k L3 8B Lunaris

L3 8B Lunaris is Sao10K's generalist and roleplay model, created as a TIES merge of five Llama 3 models: Meta-Llama-3-8B-Instruct, L3-8B-sunfall-v0.1, Jamet-8B-L3-MK1, badger-iota-llama-3-8b, and Stheno 3.2 Beta. The 8 billion parameter blend aims to combine creative dialogue with logical consistency by drawing on the complementary strengths of its components. It handles text generation and conversation with particular strength in roleplay, creative storytelling, and multi-turn dialogue, holding character consistency across multiple personas. Sao10K considers it an improvement over Stheno v3.2 for its balance of creativity and logic. Recommended settings are temperature 1.4 and min_p 0.1 with the Llama 3 Instruct template. It works well as a capable middle ground between small utility models and 70B variants for deployments where versatile performance matters more than raw size.