LyraNovaHeart/Stellar-Odyssey-12b-v0.0

Stellar Odyssey 12b v0.0: Roleplay Optimized Model Merge

Stellar-Odyssey-12b-v0.0 is a 12 billion parameter language model developed by LyraNovaHeart, specifically engineered for roleplay scenarios. This model is a sophisticated merge, utilizing the della_linear method with mistralai/Mistral-Nemo-Base-2407 as its foundational base.

Key Characteristics & Merge Details

• Architecture: A merge of several 12B models, including Sao10K/MN-12B-Lyra-v4, nothingiisreal/MN-12B-Starcannon-v2, and Gryphe/Pantheon-RP-1.5-12b-Nemo.
• Merge Method: Employs the della_linear merge technique, which is detailed in academic papers like arXiv:2406.11617 and arXiv:2212.04089.
• Instruction Format: Designed to work with the ChatML instruction format, facilitating structured conversational interactions.
• Context Length: Supports a substantial context window of 32768 tokens, allowing for extended and coherent roleplay narratives.
• Data: The model's capabilities are derived from the datasets used to train its constituent merged models, rather than new data added during the merge process.

Intended Use Case

This model is explicitly created and optimized for roleplay applications, aiming to provide engaging and consistent character interactions. Its license (CC-BY-NC-4.0) restricts commercial use without explicit permission, specifically prohibiting use by "AI Girlfriend" apps or predatory services.