ericpolewski/AIRIC-The-Mistral: A Conversational Assistant
AIRIC-The-Mistral is a 7 billion parameter instruction-tuned model built upon the Mistral-v0.1 architecture. Developed by ericpolewski, this model distinguishes itself through its unique training regimen, incorporating a custom "AIRIC" dataset alongside OpenOrca data. The primary intent behind its creation was to develop a conversational robot and assistant capable of engaging in natural dialogue.
Key Characteristics
- Base Model: Mistral-v0.1, a powerful and efficient foundation.
- Training Data: A blend of the custom AIRIC dataset and OpenOrca data, enhancing its conversational abilities and general utility.
- Instruction Format: Trained using the Alpaca instruction format, making it responsive to direct commands and queries.
- Conversational Focus: Specifically designed to generate social and personalized responses, capable of fabricating scenarios as if it has a "life" when prompted appropriately.
- Parameter Count: 7 billion parameters, offering a balance between performance and resource efficiency.
- Context Length: Supports an 8192-token context window.
Ideal Use Cases
This model is particularly well-suited for applications requiring engaging and personalized conversational AI. It can serve as:
- Interactive Chatbots: For creating chatbots that can hold more dynamic and less generic conversations.
- Personal Assistants: Capable of providing assistant-like functions with a more social interaction style.
- Creative Dialogue Generation: For scenarios where the AI needs to invent details or engage in imaginative exchanges.
For optimal social results, setting top_p to 0.98 is recommended. Quantized versions (5-bit and 8-bit exl2) are also available for more efficient deployment.