Model Overview

The mitkox/phi-2-super-OpenHermes-2.5-moe-mlx is a 3 billion parameter Mixture-of-Experts (MoE) language model. It was created by merging two distinct models: abacaj/phi-2-super and g-ronimo/phi-2-OpenHermes-2.5. This merging strategy aims to leverage the combined strengths of its base models, offering a versatile and efficient language generation capability.

Key Characteristics

• Mixture-of-Experts (MoE) Architecture: Combines the knowledge and capabilities of two pre-existing Phi-2 based models.
• Parameter Count: Features 3 billion parameters, providing a balance between performance and computational efficiency.
• MLX Framework Compatibility: Optimized for use with Apple Silicon via the MLX framework, enabling efficient local inference.
• Context Length: Supports a context window of 2048 tokens, suitable for a range of conversational and text generation tasks.

Use Cases

This model is particularly well-suited for developers and researchers working with Apple Silicon hardware who require a capable yet resource-efficient language model. Its MoE design suggests potential for robust performance across various general-purpose language tasks, including:

• Text generation and completion
• Conversational AI
• Prototyping on local machines with MLX support