0xA50C1A1/Llama-3.3-8B-Nymphaea-RP
Llama-3.3-8B-Nymphaea-RP is an 8 billion parameter language model developed by 0xA50C1A1, fine-tuned from the Llama 3.3 base for enhanced roleplay and creative writing capabilities. This model is specifically optimized for generating uncensored, diverse narrative content, making it suitable for applications requiring flexible and imaginative text generation. It features a custom mixed-quant GGUF for efficient deployment on hardware with 8/12GB VRAM.
Loading preview...
Overview
Llama-3.3-8B-Nymphaea-RP is an 8 billion parameter model, fine-tuned by 0xA50C1A1 from the Llama 3.3 base, specifically for roleplay and creative writing. The model's base weights were processed with Heretic for censorship removal, resulting in an uncensored output.
Key Capabilities
- Optimized for Roleplay and Creative Writing: Fine-tuned on an expanded Darkmere dataset, featuring diverse genres from manually curated synthetics and human-written stories.
- Uncensored Content Generation: Utilizes the Heretic tool to remove base model censorship, allowing for broader creative expression.
- Efficient Deployment: Provides a custom mixed-quant GGUF (Q5_Mix) optimized for systems with 8/12GB VRAM, with the imatrix file available for custom quantizations.
- Merge-Friendly: Designed to be easily merged with other Llama 3.1/3.3 8B fine-tunes.
Training Details
The model was trained using the DoRA (Weight-Decomposed LoRA) method with all-linear target modules, a LoRA Rank of 64, and LoRA Alpha of 64. Training involved 2 epochs with a learning rate of 1e-4 and neftune_noise_alpha=5 for noise injection.
Good For
- Generating diverse and imaginative narratives.
- Applications requiring uncensored text output.
- Developers looking for a Llama 3.3 8B fine-tune optimized for roleplay.
- Merging with other Llama 3.1/3.3 8B fine-tunes for specialized applications.