Dolphin 2.0 Mistral 7B Overview

Dolphin-2.0-mistral-7b is a 7 billion parameter language model developed by Eric Hartford, built upon the MistralAI architecture. This model is a fine-tuned version of the Dolphin dataset, which is an open-source adaptation of Microsoft's Orca, enhanced with the Airoboros dataset to boost creative capabilities. It was trained over 48 hours on 4x A100 GPUs.

Key Characteristics & Capabilities

• Uncensored and Highly Compliant: The model is intentionally uncensored, with its dataset filtered to remove alignment and bias, making it highly compliant to user requests, including potentially unethical ones. Users are advised to implement their own alignment layer for responsible deployment.
• Commercial Use: Based on MistralAI, it is suitable for both commercial and non-commercial applications.
• Prompt Format: Utilizes the ChatML prompt format for consistent interaction.

Performance Highlights

Evaluated on the Open LLM Leaderboard, Dolphin-2.0-mistral-7b demonstrates solid performance for its size:

• Average Score: 58.58
• HellaSwag (10-shot): 80.26
• Winogrande (5-shot): 75.37
• MMLU (5-shot): 56.90

Good For

• Developers seeking a highly compliant, uncensored base model for various applications.
• Use cases requiring a model that can be easily aligned to specific ethical guidelines by the user.
• Applications benefiting from enhanced creativity due to the inclusion of the Airoboros dataset.