Model Overview

Weyaxi/MetaMath-OpenHermes-2.5-neural-chat-v3-3-Slerp is a 7 billion parameter language model developed by Weyaxi. This model is a result of a strategic merge using the slerp (spherical linear interpolation) method, combining two distinct base models: meta-math/MetaMath-Mistral-7B and PulsarAI/OpenHermes-2.5-neural-chat-v3-3-Slerp. The merging process specifically targets different layers for interpolation, with self_attn layers and mlp layers having varied t values, and a fallback value for other tensors.

Key Capabilities

• Hybrid Performance: Integrates the specialized mathematical reasoning abilities from MetaMath-Mistral-7B with the strong conversational and instruction-following capabilities of OpenHermes-2.5-neural-chat-v3-3-Slerp.
• Merged Architecture: Utilizes mergekit with a slerp merge method, allowing for a nuanced combination of features from its constituent models.
• Mistral Base: Built upon the Mistral-7B-v0.1 architecture, providing a solid foundation for performance.

Good For

• Applications requiring a balance between mathematical problem-solving and general-purpose dialogue generation.
• Use cases where a single model needs to handle both complex numerical reasoning and natural language interaction effectively.