GPT-NeoX-20B-Erebus
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Framework | PyTorch |
| Paper | Research Paper |
| Training Infrastructure | TPUv3-256 TPU pod |
What is GPT-NeoX-20B-Erebus?
GPT-NeoX-20B-Erebus is a specialized language model derived from the GPT-NeoX-20B architecture, specifically fine-tuned for adult content generation. The model name "Erebus," derived from Greek mythology meaning "darkness," aligns with its predecessor's theme of "deep abyss." It represents the second generation of the original Shinen model, developed by Mr. Seeker and the KoboldAI community.
Implementation Details
The model was trained using a heavily modified version of Ben Wang's Mesh Transformer JAX library on a TPUv3-256 TPU pod. The training process incorporated six distinct datasets, all focused on adult-themed content, with specific quality thresholds for inclusion. The model implements a tagging system using the format [Genre:
- Utilizes the GPT-NeoX architecture with 20B parameters
- Trained on carefully curated high-quality content from multiple sources
- Implements specialized genre tagging system
- Built with PyTorch framework
Core Capabilities
- Advanced text generation with adult-themed content focus
- Genre-aware content generation
- High-quality output based on highly-rated source material
- Specialized content handling with genre tagging support
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its specialized training on carefully curated adult content from six different high-quality sources, combined with its implementation of the powerful GPT-NeoX architecture. It's specifically designed for mature content generation with a sophisticated understanding of genres and themes.
Q: What are the recommended use cases?
The model is strictly designed for adult content generation and should only be used by individuals over 18 years of age. It's particularly suited for creative writing and content generation within its specific domain, with strong content warnings and usage restrictions in place.
