CNC-7b is a lora adapter for Mistral-7b (Instruct) intended to be clear, concise, and helpful in short text conversations. It is designed for conversational agents and assistants.
Training Data
CNC-7b was trained on synthetic conversational data generated by Newstar using ChatGPT. The data was shaped using custom instructions to encourage clear, concise, and helpful responses.
Evaluation Data
CNC-7b was evaluated on a test set of human-human conversations to measure whether responses were clear, concise, and on-topic.
Ethical Considerations
CNC-7b has limited conversational abilities and is not intended for complex conversations.
The training data was filtered to remove harmful, unethical, or dangerous content.
The model has no notion of facts about the real world. Any factual statements generated should not be assumed to be true.
Caveats and Recommendations
Only the Peft adapter parameters are released for CNC-7b. The full model is not released.
CNC-7b has limited knowledge outside of conversational abilities. Do not use for anything requiring real world knowledge.
Monitor CNC-7b conversations for harmful content generated, and re-train the model as needed.
