knifeayumu/Negative-Anubis-70B-v1

Model Overview

Negative-Anubis-70B-v1 is a 70 billion parameter language model created by knifeayumu through a strategic merge of two distinct models: SicariusSicariiStuff/Negative_LLAMA_70B and TheDrummer/Anubis-70B-v1. The primary motivation behind this merge was to enhance the prose style, specifically to introduce more verbosity and to counteract the inherent positivity bias observed in Anubis-70B-v1, while retaining the 'negative' characteristics of Negative_LLAMA_70B.

Merge Details

This model was constructed using the SLERP merge method via mergekit. The configuration involved specific weighting parameters (t: [0.1, 0.55, 1, 0.55, 0.1]) applied to the constituent models, with TheDrummer/Anubis-70B-v1 serving as the base model. The merge was performed using bfloat16 dtype.

Key Characteristics

• Balanced Tone: Aims to provide a more balanced output by mitigating strong positivity biases while avoiding overly dry or negative prose.
• Enhanced Verbosity: Designed to produce more descriptive and verbose text compared to its 'Negative_LLAMA' component.

Potential Use Cases

This model is suitable for applications where a nuanced, less overtly positive, and more verbose textual output is desired, particularly in scenarios where the original components' individual biases or stylistic limitations were a concern.