Heralax/mannerstral-7b
Heralax/mannerstral-7b is a 7 billion parameter, Mistral-derived causal language model fine-tuned by Heralax. It specializes in providing factual answers related to manners and etiquette, particularly from the previous century, leveraging data generated from Project Gutenberg. The model is optimized for tightly focused factual question answering within its domain, with a context length of 4096 tokens.
Loading preview...
Mannerstral 7b Overview
Mannerstral 7b is a 7 billion parameter language model developed by Heralax, specifically fine-tuned to be a domain expert on manners and etiquette. This model is built upon a Mistral-derived architecture and utilizes a 4096-token context length. Its training data, generated with Augmentoolkit using Llama 3 70b and Llama 3 8b, is heavily focused on historical etiquette, primarily sourced from Project Gutenberg.
Key Capabilities
- Domain Expertise: Highly specialized in factual question answering concerning manners and etiquette, particularly from the previous century.
- Factual QA: Designed for tightly focused factual responses within its domain.
- ChatML Support: Utilizes the ChatML format for conversations.
- Low Temperature Recommended: Optimal performance is achieved with low inference temperatures (e.g., 0).
Good For
- Historical Etiquette Research: Ideal for applications requiring knowledge of past social customs and proper conduct.
- Factual Information Retrieval: Suitable for scenarios where precise, domain-specific answers on etiquette are needed.
- Specialized Chatbots: Can serve as the core for chatbots focused on historical or formal social interactions.
Model Quirks
While capable, the model is subject to leading questions; it may not correct a user if a follow-up question deviates from polite discourse, even if the initial query was about etiquette. Prompting strategies may help mitigate this behavior. It does not include generalist assistant data, though it shows some general capabilities.