| Property | Value |
|---|---|
| Base Architecture | LLaMA 30B |
| Framework | PyTorch |
| Primary Use Cases | Text Generation, Instruction Following, Storytelling |
What is 30B-Lazarus?
30B-Lazarus is an experimental language model that pushes the boundaries of LoRA application and model merging techniques. Created by CalderaAI, it represents a sophisticated combination of multiple high-performing models and LoRAs, including SuperCOT, GPT4-Alpaca, and Vicuna components.
Implementation Details
The model employs a complex architecture that layers multiple LoRAs on a composite model base. The core composition follows the pattern [SuperCOT([gtp4xalpaca(manticorechatpygalpha+vicunaunlocked)]+[StoryV2(kaiokendev-SuperHOT-LoRA-prototype30b-8192)])].
- Built on LLaMA 30B architecture
- Implements multiple LoRA layers for enhanced capabilities
- Combines storytelling and instruction-following abilities
- Supports both Alpaca and Vicuna instruction formats
Core Capabilities
- Advanced text generation and storytelling
- Enhanced instruction following
- Uncensored output capability
- Flexible response generation with adjustable parameters
- Support for multiple instruction formats
Frequently Asked Questions
Q: What makes this model unique?
The model's unique approach to combining multiple LoRAs and model merges in a way that aims to preserve and enhance desired features without diluting the model's effectiveness makes it stand out. It's specifically designed to maintain the strengths of each component while minimizing interference between different LoRAs.
Q: What are the recommended use cases?
The model is particularly well-suited for storytelling and instruction-following tasks. It works best with Alpaca instruct format, though it also supports Vicuna format. Users are recommended to experiment with different presets (particularly Godlike and Storywriter) and adjust output length and temperature settings for optimal results.