Overview

ABX-AI/Silver-Sun-11B is a 10.7 billion parameter language model resulting from a SLERP merge of ABX-AI/Solstice-FKL-11B (itself a mixture of Sao10K/Solstice-11B-v1 and saishf/Fimbulvetr-Kuro-Lotus-10.7B) and Himitsui/Kaiju-11B. This model is designed for writing and reasoning, exhibiting a distinct preference for uncensored and profane NSFW language when specific prompts are used. It is recommended for use with Alpaca instruction presets.

Key Capabilities

• Writing and Reasoning: Demonstrates proficiency in generating text and performing reasoning tasks.
• Uncensored Output: Known for its highly uncensored nature, including a preference for profane NSFW language under appropriate conditions.
• Merge Architecture: Utilizes the SLERP merge method, combining strengths from its constituent models.

Performance Insights

OpenLLM Eval results indicate an average score of 74.13, with individual metrics including:

• AI2 Reasoning Challenge (25-Shot): 69.80
• HellaSwag (10-Shot): 87.91
• MMLU (5-Shot): 66.90
• TruthfulQA (0-shot): 61.89
• Winogrande (5-shot): 84.14

Note that the GSM8K score was excluded from the average due to formatting issues, though practical usage does not reflect this problem. The model's performance is considered comparable to models like upstage/SOLAR-10.7B-v1.0.

Good For

• Applications requiring direct, unfiltered, or uncensored language generation.
• Creative writing or roleplay scenarios where profane or NSFW content is desired.
• Experimentation with merged 11B parameter models for specific linguistic styles.