vicgalle/Roleplay-Llama-3-8B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Apr 19, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

The vicgalle/Roleplay-Llama-3-8B is an 8 billion parameter Llama-3 model fine-tuned for roleplay scenarios, specifically generating dialogue with interspersed actions. It was trained on the NSFW_RP_Format_DPO dataset, resulting in a distinct output format. This model excels in interactive narrative generation and achieved the top performance for its parameter size on the Chaiverse leaderboard. With an 8192-token context length, it is optimized for engaging and formatted roleplay interactions.

Loading preview...

Model Overview

The vicgalle/Roleplay-Llama-3-8B is an 8 billion parameter model based on the Llama-3 architecture, specifically fine-tuned for roleplay generation. Its training utilized the ResplendentAI/NSFW_RP_Format_DPO dataset, which conditions the model to produce outputs in a specific format: dialogue *action*.

Key Capabilities

  • Formatted Roleplay Generation: Designed to generate conversational text interspersed with actions, ideal for interactive storytelling and character-driven scenarios.
  • High Performance in Roleplay: Achieved the second-highest ELO score on the Chaiverse leaderboard as of April 23, 2024, and is noted as the best-performing 8B parameter model in this category.
  • Llama-3 Base: Benefits from the robust capabilities of the Llama-3 foundation model.

Evaluation Highlights

While specialized for roleplay, the model's general LLM capabilities were evaluated on the Open LLM Leaderboard, showing an average score of 24.33. Specific metrics include:

  • IFEval (0-Shot): 73.20
  • BBH (3-Shot): 28.55
  • MMLU-PRO (5-shot): 30.09

Good For

  • Interactive Storytelling: Creating dynamic and engaging narrative experiences.
  • Character Simulation: Generating responses that include both dialogue and character actions.
  • Roleplay Applications: Any use case requiring structured, action-oriented conversational outputs.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p