PYAE1994/Roleplay-Llama-3-8B
PYAE1994/Roleplay-Llama-3-8B is an 8 billion parameter Llama-3 model fine-tuned specifically for roleplay scenarios, generating dialogue interspersed with actions. It was trained using the ResplendentAI/NSFW_RP_Format_DPO dataset to produce structured roleplay outputs. This model excels in interactive narrative generation, achieving a high ranking on the Chaiverse leaderboard for its parameter size. Its primary strength lies in creating dynamic and formatted roleplay conversations.
Loading preview...
Overview
PYAE1994/Roleplay-Llama-3-8B is an 8 billion parameter Llama-3 model that has been fine-tuned for roleplay generation. Its training utilized the ResplendentAI/NSFW_RP_Format_DPO dataset, which specifically teaches the model to format its outputs with dialogue and interspersed actions (e.g., dialogue *action*).
Key Capabilities
- Structured Roleplay Generation: Produces responses formatted with dialogue and embedded actions, ideal for interactive narrative experiences.
- High Performance in Roleplay: Achieves a strong ranking on the Chaiverse leaderboard, notably being the top-performing 8B parameter model by ELO score as of April 23, 2024.
- Llama-3 Architecture: Benefits from the foundational capabilities of the Llama-3 model family.
Evaluations
While optimized for roleplay, the model's general language understanding and reasoning metrics from the Open LLM Leaderboard include:
- Avg. Score: 24.33
- IFEval (0-Shot): 73.20
- BBH (3-Shot): 28.55
- MMLU-PRO (5-shot): 30.09
Good For
- Applications requiring dynamic and formatted roleplay interactions.
- Generating character dialogue with integrated actions for storytelling or interactive fiction.
- Use cases where a compact yet capable roleplay-specific model is needed.