arcee-ai/SEC-1.6-MBX-7B-DPO
SEC-1.6-MBX-7B-DPO is a 7 billion parameter language model developed by arcee-ai, created by merging arcee-ai/sec-mistral-7b-instruct-1.6-epoch and macadeliccc/MBX-7B-v3-DPO. This model leverages a slerp merge method to combine the strengths of its base models, offering a balanced performance profile. It is designed for general-purpose language tasks, benefiting from the instruction-tuned and DPO-optimized characteristics of its merged components.
Loading preview...
Model Overview
SEC-1.6-MBX-7B-DPO is a 7 billion parameter language model developed by arcee-ai. It is a product of merging two distinct models: arcee-ai/sec-mistral-7b-instruct-1.6-epoch and macadeliccc/MBX-7B-v3-DPO. This merge was performed using the mergekit tool, specifically employing a slerp (spherical linear interpolation) method to combine the weights of the constituent models.
Key Characteristics
- Merged Architecture: Combines an instruction-tuned model (
sec-mistral-7b-instruct-1.6-epoch) with a DPO-optimized model (MBX-7B-v3-DPO). - Slerp Merge Method: Utilizes spherical linear interpolation for a smooth and effective combination of model parameters.
- Parameter Configuration: Specific
tvalues were applied during the merge, with varying weights forself_attnandmlplayers, indicating a fine-tuned approach to balancing the contributions of each base model. - Bfloat16 Precision: The merged model is configured to use
bfloat16data type, optimizing for both performance and memory efficiency.
Intended Use Cases
This model is suitable for a variety of general language generation and understanding tasks, benefiting from the instruction-following capabilities of its base models. Its DPO optimization suggests improved alignment with human preferences, making it potentially useful for applications requiring nuanced and coherent responses.