Novaciano/Hereticsutra-2B
Novaciano/Hereticsutra-2B is a 2.6 billion parameter language model with an 8192-token context length, created by Novaciano through a SLERP merge of ChiKoi7/Gemma-2-Llama-Swallow-2b-it-v0.1-Heretic and TheDrummer/Gemmasutra-Mini-2B-v1. This model is characterized by its unique merge configuration, which introduces behavioral instability and degraded alignment while preserving host cognition, making it suitable for experimental applications exploring altered model behaviors.
Loading preview...
Novaciano/Hereticsutra-2B: Merged Language Model
Novaciano/Hereticsutra-2B is a 2.6 billion parameter language model resulting from a SLERP merge of two pre-trained models: ChiKoi7/Gemma-2-Llama-Swallow-2b-it-v0.1-Heretic and TheDrummer/Gemmasutra-Mini-2B-v1. This merge was performed using mergekit with a specific configuration designed to alter the model's characteristics.
Key Characteristics
- Merge Method: Utilizes the Spherical Linear Interpolation (SLERP) method for combining model weights.
- Parameter Configuration: The merge applies a
tfactor of 0.45, indicating a partial influence from the secondary model. - Behavioral Impact: The configuration explicitly notes that this merge aims to preserve host cognition while introducing degraded alignment and behavioral instability.
Potential Use Cases
This model is particularly suited for research and experimental applications where the exploration of altered or unconventional model behaviors is desired. Its unique merge parameters suggest it could be used for:
- Investigating the effects of specific weight interpolations on model alignment and stability.
- Developing applications that require models with intentionally modified or 'heretical' response patterns.
- Exploring the boundaries of model control and predictability in fine-tuned or merged architectures.