jeonsworld/CarbonVillain-en-10.7B-v1 is a 10.7 billion parameter experimental language model created by jeonsworld using mergekit, specifically designed to oppose indiscriminate carbon emissions. This model was developed by merging Weyaxi/SauerkrautLM-UNA-SOLAR-Instruct and VAGOsolutions/SauerkrautLM-SOLAR-Instruct using the slerp method. It is intended for use in scenarios where a model with a specific ethical stance against carbon emissions is desired, offering a unique perspective compared to general-purpose LLMs.
Loading preview...
Overview
jeonsworld/CarbonVillain-en-10.7B-v1 is an experimental 10.7 billion parameter language model developed by jeonsworld. Its core design principle is to oppose indiscriminate carbon emissions, setting it apart from conventional LLMs. The model was constructed using mergekit by combining two existing models: Weyaxi/SauerkrautLM-UNA-SOLAR-Instruct and VAGOsolutions/SauerkrautLM-SOLAR-Instruct, utilizing the slerp merging method.
Key Characteristics
- Ethical Stance: Explicitly designed with a focus on opposing carbon emissions.
- Architecture: A merged model, combining the strengths of its base components.
- Parameter Count: 10.7 billion parameters, offering a balance between capability and computational requirements.
- Prompt Template: Uses a simple
### User: {user}\n\n### Assistant: {asistant}format.
Performance Benchmarks
Based on the Open LLM Leaderboard, CarbonVillain-en-10.7B-v1 achieves an average score of 74.28 across various tasks. Notable scores include:
- ARC (25-shot): 71.24
- HellaSwag (10-shot): 88.45
- MMLU (5-shot): 66.42
- TruthfulQA (0-shot): 71.97
- Winogrande (5-shot): 83.26
- GSM8K (5-shot): 64.29
Good For
- Applications requiring a model with an inherent bias against carbon emissions.
- Research into model merging techniques and their impact on ethical alignment.
- Exploring the capabilities of models derived from specific base LLMs like SauerkrautLM-UNA-SOLAR-Instruct and SauerkrautLM-SOLAR-Instruct.