Overview

jeonsworld/CarbonVillain-en-10.7B-v4 is a 10.7 billion parameter experimental language model created by jeonsworld, specifically developed to oppose indiscriminate carbon emissions. This version is a result of merging previous iterations, CarbonVillain-en-10.7B-v1 and v2, using the slerp method via mergekit.

Key Capabilities & Performance

The model demonstrates strong performance across several benchmarks, achieving an average score of 74.52 on the Open LLM Leaderboard. Notable scores include:

• HellaSwag (10-Shot): 88.48
• Winogrande (5-Shot): 83.58
• AI2 Reasoning Challenge (25-Shot): 71.25
• TruthfulQA (0-shot): 71.95
• MMLU (5-Shot): 66.27
• GSM8k (5-shot): 65.58

Prompt Template

Users should interact with the model using the following prompt format:

### User:
{user}

### Assistant:
{asistant}

Intended Use

This model is suitable for applications requiring a language model with a specific focus on environmental themes, particularly those related to carbon emissions, while maintaining solid general language understanding and reasoning capabilities as indicated by its benchmark performance.