shrugging-shoulders/Amberlight-12B
Amberlight-12B is a 12 billion parameter language model developed by shrugging-shoulders, currently a work-in-progress fine-tune focused on roleplay (RP) scenarios. It features decent instruction following, multilingual capabilities, and good writing quality, with a context length of 32768 tokens. The model is designed to be uncensored for NSFW content when prompted, making it suitable for creative and interactive storytelling applications.
Loading preview...
Amberlight-12B: A Roleplay-Focused Language Model
Amberlight-12B, developed by shrugging-shoulders, is a 12 billion parameter language model currently undergoing fine-tuning specifically for roleplay (RP) applications. This model is noted for its developing capabilities in instruction following, multilingual support, and overall writing quality, making it a candidate for interactive narrative generation.
Key Characteristics & Capabilities
- Roleplay Optimization: The primary focus of this model is on roleplay scenarios, aiming for engaging and responsive interactions.
- Uncensored Content: It is designed to handle NSFW content without refusal, provided it is appropriately prompted.
- Multilingual Support: The model possesses multilingual capabilities, broadening its applicability across different language contexts.
- Writing Quality: Users can expect a generally good quality of written output from the model.
- Context Length: Supports a substantial context window of 32768 tokens, allowing for longer and more complex interactions.
Current Status and Known Issues
Amberlight-12B is described as "60% cooked" and is still under active development. Known issues include a kinetic storytelling style, potential pacing jumps, and occasional imperfections in output quality. Future plans involve additional rounds of supervised fine-tuning (SFT) and extensive DPO (Direct Preference Optimization) to refine its performance.
Recommended Usage
For optimal performance, users are advised to use ChatML formatting and experiment with inference parameters such as temperature (0.8-1), top-p (0.95 with min-p 0.025, or 0.90 with min-p 0.05).