CollectiveCognition-v1.1-Nebula-7B Overview

CollectiveCognition-v1.1-Nebula-7B is a 7 billion parameter language model resulting from a merge of two distinct models: teknium/CollectiveCognition-v1.1-Mistral-7B and PulsarAI/Nebula-7B. This integration aims to leverage the strengths of both base models, providing a versatile tool for various natural language processing applications. The model supports an 8192-token context length, allowing for processing longer inputs and generating more coherent, extended outputs.

Performance Highlights

Evaluated on the Open LLM Leaderboard, CollectiveCognition-v1.1-Nebula-7B achieves an average score of 53.79. Key benchmark results include:

• ARC (25-shot): 58.11
• HellaSwag (10-shot): 82.39
• MMLU (5-shot): 57.03
• TruthfulQA (0-shot): 53.53
• Winogrande (5-shot): 73.72
• GSM8K (5-shot): 9.55
• DROP (3-shot): 42.17

These scores indicate a balanced capability across reasoning, common sense, and factual recall tasks, making it a solid choice for general-purpose applications.

Good For

• General text generation: Creating coherent and contextually relevant text.
• Question answering: Responding to queries based on provided context or general knowledge.
• Reasoning tasks: Handling tasks that require logical inference, as indicated by its ARC and MMLU scores.
• Common sense understanding: Demonstrating good performance on HellaSwag and Winogrande benchmarks.