| Property | Value |
|---|---|
| Base Model | ToastyPigeon/ms3-roselily-instruct |
| Parameter Count | 24B |
| Model URL | Hugging Face |
| Author | allura-org |
What is MS3-24B-Roselily-Creative?
MS3-24B-Roselily-Creative is a specialized language model fine-tuned for creative writing, roleplaying, and storytelling applications. Built upon the MS3-roselily-instruct base model, it has been enhanced with creative writing datasets while maintaining compatibility with multiple instruction formats.
Implementation Details
The model implements multiple instruction formats including ChatML and Alpaca, with special consideration for Tekken v7 compatibility. It features custom token assignments (20 and 21) for ChatML while preserving Tekken tokens, enabling potential merges with Tekken models.
- Supports multiple instruction formats: ChatML, Alpaca, and Tekken v7
- Custom token assignments for format compatibility
- Optimized for creative generation tasks
- Less sensitive to sampling parameters compared to standard instruct models
Core Capabilities
- Creative writing and storytelling
- Roleplay interactions
- Flexible instruction following
- Chat and adventure mode completion
- Multi-format support with proper generation handling
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its optimization for creative tasks while maintaining multiple instruction format compatibility. It combines the robustness of MS3 architecture with specialized creative writing capabilities.
Q: What are the recommended use cases?
This model is best suited for creative writing, storytelling, roleplay scenarios, and interactive narrative generation. It works particularly well with conservative sampling parameters for optimal output quality.