Model Overview
SEC-MBX-7B-DPO is a 7 billion parameter language model developed by arcee-ai, created through a merge of two distinct models using mergekit. This model combines the capabilities of:
- arcee-ai/sec-mistral-7b-instruct-1.2-epoch: An instruction-tuned Mistral-based model.
- macadeliccc/MBX-7B-v3-DPO: A model likely optimized using Direct Preference Optimization (DPO).
The merge process utilized a slerp (spherical linear interpolation) method, with specific t parameters applied to different architectural components like self_attn and mlp layers, indicating a fine-tuned approach to blending the source models' characteristics. The base model for this merge was arcee-ai/sec-mistral-7b-instruct-1.2-epoch, and the model operates in bfloat16 precision.
Key Characteristics
- Architecture: Based on the Mistral architecture, providing a strong foundation for language tasks.
- Parameter Count: 7 billion parameters, offering a balance between performance and computational efficiency.
- Context Length: Supports a context window of 4096 tokens.
- Merge Method: Employs
slerp for combining models, allowing for nuanced integration of features.
Intended Use Cases
This model is suitable for a variety of general-purpose natural language processing tasks, benefiting from the instruction-tuning and DPO optimization of its merged components. It can be applied to areas requiring robust language understanding and generation.