Synthia-7B-v3.0: Enhanced Instruction Following and Reasoning

Synthia-7B-v3.0 is a 7 billion parameter model built upon the Mistral-7B architecture, developed by migtissera. It has been meticulously fine-tuned to improve instruction following and facilitate long-form conversations, drawing inspiration from the Orca-2 paper's methodologies.

Key Capabilities & Features

• Advanced Reasoning: Incorporates a Generalized Tree-of-Thought prompt within its training dataset, designed to evoke structured reasoning and backtracking for cohesive thought chains.
• Instruction Following: Optimized for accurately understanding and executing complex instructions.
• Long-Form Conversations: Excels in maintaining coherent and extended dialogues, making it suitable for interactive applications.
• System Context Integration: While trained without system context as per Orca-2 suggestions, it can be prompted with a specific system message to activate Tree of Thought + Chain of Thought reasoning.

Usage Considerations

To leverage its advanced reasoning capabilities, users can employ a system message such as: "Elaborate on the topic using a Tree of Thoughts and backtrack when necessary to construct a clear, cohesive Chain of Thought reasoning. Always answer without hesitation."

Limitations

As an uncensored model, Synthia-7B-v3.0 may occasionally produce inaccurate, inappropriate, biased, or offensive content. Users should exercise caution and cross-check information as needed.