grimjim/Magnolia-v5a-12B
grimjim/Magnolia-v5a-12B is a 12.2 billion parameter language model created by grimjim using a Task Arithmetic merge of several pre-trained models, including grimjim/Magnolia-v3-12B, Nitral-AI/Captain_BMO-12B, and natong19/Mistral-Nemo-Instruct-2407-abliterated. It leverages Nemo 12B Base as its foundation for the merge, incorporating models fine-tuned with instruct and narrative generation data. This model is designed for general language tasks, benefiting from the combined strengths of its merged components.
Loading preview...
Overview
grimjim/Magnolia-v5a-12B is a 12.2 billion parameter language model developed by grimjim, created through a Task Arithmetic merge using mergekit. The base model for this merge was grimjim/mistralai-Mistral-Nemo-Base-2407, which is a variant of Nemo 12B Base. This approach combines the strengths of multiple models to enhance overall performance.
Key Capabilities
- Merged Architecture: Utilizes a Task Arithmetic merge method, integrating diverse model capabilities.
- Base Model: Built upon the Nemo 12B Base, providing a strong foundation for language understanding and generation.
- Component Integration: Incorporates models fine-tuned with a mix of instruct and narrative generation data, alongside an 'abliterated' Instruct model that showed strong performance.
Good for
- General Language Tasks: Suitable for a broad range of applications benefiting from combined instruct and narrative generation capabilities.
- Experimentation with Merged Models: Ideal for researchers and developers interested in exploring the performance characteristics of models created via Task Arithmetic merging.