Walmart-the-bag/MysticFusion-13B
MysticFusion-13B by Walmart-the-bag is a 13 billion parameter language model with a 4096-token context length, created by linearly merging KoboldAI/LLaMA2-13B-Tiefighter, NeverSleep/Noromaid-13b-v0.1.1, and lmsys/vicuna-13b-v1.5. This model is specifically designed and optimized for story writing and basic instruction following. It achieves an average score of 59.31 on the Open LLM Leaderboard, with notable performance in HellaSwag (84.43) and Winogrande (76.01).
Loading preview...
MysticFusion-13B: A Merged Model for Creative Writing
MysticFusion-13B is a 13 billion parameter language model developed by Walmart-the-bag, built through a linear merge of three distinct 13B models: KoboldAI/LLaMA2-13B-Tiefighter (30% weight), NeverSleep/Noromaid-13b-v0.1.1 (50% weight), and lmsys/vicuna-13b-v1.5 (20% weight). This fusion aims to combine the strengths of its constituent models, primarily focusing on generative text tasks.
Key Capabilities & Performance
- Optimized for Story Writing: The model's primary design goal is to excel in creative story generation, making it suitable for narrative-focused applications.
- Basic Instruction Following: Beyond creative tasks, it also supports fundamental instruction-based interactions.
- Open LLM Leaderboard Performance: MysticFusion-13B achieved an average score of 59.31 on the Hugging Face Open LLM Leaderboard. Specific benchmark results include:
- HellaSwag (10-Shot): 84.43
- Winogrande (5-shot): 76.01
- AI2 Reasoning Challenge (25-Shot): 61.35
- MMLU (5-Shot): 57.29
- TruthfulQA (0-shot): 51.98
- GSM8k (5-shot): 24.79
Usage & Prompting
The model is intended for applications requiring creative text generation, particularly story writing. It utilizes the Alpaca prompt template for instruction-based interactions:
### Instruction:
### Response:
When to Consider Using MysticFusion-13B
- Creative Content Generation: Ideal for tasks like generating fictional stories, character dialogues, or descriptive passages.
- Experimentation with Merged Models: Developers interested in exploring the capabilities of models created via linear merging techniques.
- Applications requiring a balance of creativity and basic instruction following.