Model Overview
FunkEngine/SchweinZwei-13b is an instruction-tuned language model based on Meta AI's Llama-2 13B architecture. Developed by FunkEngine, this model (formerly known as Metharme) is specifically designed for conversational, roleplaying, and storywriting applications. It aims to provide a model that can be guided using natural language, similar to other instruct models, but with a strong bias towards creative and fictional content generation.
Key Capabilities
- Fiction Writing: Optimized for generating fictional stories and narrative content.
- Roleplaying: Excels in maintaining character personas and generating engaging roleplay dialogues.
- Conversational AI: Capable of natural language conversations, with a focus on creative and interactive exchanges.
- Flexible Prompting: Utilizes a
<|system|>, <|user|>, <|model|> token format for structured conversations and persona injection, allowing for detailed control over model behavior and response length.
Training and Data
The model was fine-tuned using supervised learning on a diverse dataset. This dataset includes FunkEngine's proprietary PIPPA dataset, various instruction datasets, and data sourced from roleplaying forums. The training methodology focused on blending regular instruction data with roleplay scenarios, fictional narratives, and conversations augmented with synthetically generated instructions.
Intended Use and Limitations
SchweinZwei-13b is intended primarily for fictional writing and entertainment purposes. It is important to note that the model was not fine-tuned for safety or harmlessness. Due to its training on data that may contain profanity, lewd, or offensive texts, it can produce socially unacceptable or factually incorrect outputs. Users should be aware of these biases and limitations, especially when deploying the model in sensitive applications.