OpenCerebrum-2.0-7B Overview

OpenCerebrum-2.0-7B is a 7 billion parameter open-source language model, fine-tuned by Locutusque from the alpindale/Mistral-7B-v0.2-hf base. Its development focused on replicating the capabilities of Aether Research's proprietary Cerebrum model, utilizing advanced in-house curation techniques from Cognitive Computations.

Key Capabilities & Training

• Diverse Fine-Tuning: The model underwent SFT and DPO fine-tuning on approximately 7,000 examples, drawing from 15 distinct data sources. This diverse dataset covers areas such as coding, mathematics, science, multi-turn conversations, Retrieval Augmented Generation (RAG), reasoning, and general instruction-following.
• Broad Application: Designed to be a versatile tool, it aims to provide strong performance across a wide range of benchmarks relevant to complex cognitive tasks.
• Open-Source & License: Released under the Apache 2.0 license, promoting accessibility and further development within the community.

Intended Use Cases

• Coding and Scientific Tasks: Strong performance in programming, mathematical problem-solving, and scientific inquiry.
• General Instruction Following: Capable of handling a variety of general question-answering and text generation tasks.
• Reasoning and Conversation: Equipped for multi-turn conversations and complex reasoning challenges.

Limitations

While aiming for high performance, the model may exhibit biases inherited from its training data and has computational constraints typical of a 7B parameter model compared to larger alternatives. Thorough testing is recommended for specific applications.