OpenCerebrum-1.0-7B-DPO Overview

OpenCerebrum-1.0-7B-DPO is a 7 billion parameter open-source language model developed by Locutusque. It is fine-tuned from the alpindale/Mistral-7B-v0.2-hf base model, with the explicit goal of replicating the performance of Aether Research's proprietary Cerebrum model.

Key Capabilities & Training

This model was fine-tuned on a diverse dataset comprising approximately 21,000 examples across six public datasets. These datasets cover a range of domains including coding, mathematics, science, reasoning, and general instruction-following, aiming to equip the model with broad knowledge and strong reasoning abilities. The training utilized the ChatML prompt format.

Performance & Benchmarks

Evaluations on the Open LLM Leaderboard show an average score of 62.78. Specific benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 62.71
• HellaSwag (10-Shot): 84.33
• MMLU (5-Shot): 62.59
• GSM8k (5-Shot): 42.00

Intended Use Cases

OpenCerebrum-1.0-7B-DPO is designed to be a versatile model suitable for:

• Coding tasks
• Mathematical problem-solving
• Scientific inquiry and analysis
• General question-answering and text generation
• Reasoning-intensive applications

Limitations

While aiming to replicate Cerebrum's capabilities, this model's performance may not fully match the proprietary version due to its smaller fine-tuning dataset (~21,000 examples). It may also inherit biases from its training data, and its 7B parameter size implies computational and memory constraints compared to larger models.