ArliAI/Gemma-2-2B-ArliAI-RPMax-v1.1
Gemma-2-2B-ArliAI-RPMax-v1.1 is a 2.6 billion parameter language model from ArliAI, based on the Gemma-2 architecture, with an 8192 token context length. It is specifically fine-tuned for creative writing and roleplay, utilizing a diverse and deduplicated dataset to ensure high creativity and non-repetitive outputs. The model is designed to understand and appropriately act across varied characters and situations, avoiding personality latching common in other roleplay models. It aims to provide a distinct style for interactive narrative generation.
Loading preview...
Model Overview
Gemma-2-2B-ArliAI-RPMax-v1.1 is a 2.6 billion parameter model developed by ArliAI, built upon the Gemma-2-2B-it base. It is part of the RPMax series, which focuses on enhancing creative writing and roleplay capabilities across various model sizes.
Key Capabilities & Training
- Creative Roleplay: The model is specifically trained on a diverse and deduplicated dataset of creative writing and roleplay scenarios. This approach ensures the model does not repeat characters or situations, promoting high creativity and preventing it from fixating on specific personalities.
- Non-Repetitive Outputs: Designed to avoid "repetition sickness" and "in-bred" outputs often found in other roleplay models, offering a distinct and varied interaction style.
- Efficient Training: The model underwent 1 epoch of training for minimized repetition sickness, utilizing QLORA with 64-rank 128-alpha, resulting in approximately 2% trainable weights. Training was completed in under 1 day on 2x3090Ti GPUs.
- Context Length: Supports a sequence length of 4096 tokens during training, with an effective context length of 8192 tokens.
Suggested Use Cases
- Interactive Storytelling: Ideal for generating dynamic and varied narratives in roleplaying games or interactive fiction.
- Character Development: Useful for creating and maintaining distinct character personalities and interactions without repetition.
- Creative Content Generation: Suitable for applications requiring highly creative and non-formulaic text generation, particularly in conversational or narrative contexts.