Delcos/Mistral-Pygmalion-7b
Delcos/Mistral-Pygmalion-7b is a 7 billion parameter language model, merging the Mistral architecture with Pygmalion's roleplay capabilities. This model is specifically optimized for roleplaying scenarios, offering improved results while maintaining a smaller footprint. While primarily focused on roleplay, it also functions as a capable assistant, though its 7B size limits complex reasoning tasks.
Loading preview...
Delcos/Mistral-Pygmalion-7b Overview
Delcos/Mistral-Pygmalion-7b is a 7 billion parameter model developed by Delcos, created by merging the Mistral architecture with the Pygmalion model. The primary goal of this merger is to preserve and enhance roleplay abilities while benefiting from Mistral's improved performance.
Key Capabilities & Performance
This model excels particularly in roleplaying scenarios, making it a strong choice for applications requiring character interaction and narrative generation. While its 7B parameter count means it may not be ideal for highly complex reasoning tasks, it still serves as a fairly capable general assistant. Performance metrics from the LLM Leaderboard Evaluation include:
- Avg. Score: 44.58
- ARC (25-shot): 54.44
- HellaSwag (10-shot): 78.48
- MMLU (5-shot): 49.23
- TruthfulQA (0-shot): 41.82
- Winogrande (5-shot): 75.3
- GSM8K (5-shot): 6.82
- DROP (3-shot): 5.94
Prompt Template
The model utilizes a straightforward instruction-based prompt template:
### Instruction:
{Prompt & Backstory}
### Assistant:
{Output}An example of its use for character definition is provided, demonstrating how to set up a persona for the assistant. A GGUF version is also available, provided by TheBloke.