Angel Slayer 12 B Unslop Mell R P Max D A R K N E S S

Introduction

The AngelSlayer-12B-Unslop-Mell-RPMax-DARKNESS model is a merged pre-trained language model designed to counter potential positivity biases while maintaining stability. It is optimized for varied prose and creative writing, with a focus on role-playing and conversational contexts. Key features include expansive text generation capabilities and the ability to maintain character consistency through prompts.

Architecture

AngelSlayer-12B-Unslop-Mell-RPMax-DARKNESS combines several models using the della_linear merge method. The models included in this merge are:

• Inflatebot/MN-12B-Mag-Mell-R1
• ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2
• DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS
• TheDrummer/UnslopNemo-12B-v4.1 (base model)

The configuration involves weights and densities specific to each model, with a bfloat16 data type and a chatml chat template.

Training

The model is currently in the testing stage, evaluating its ability to handle context and maintain character consistency. It addresses issues like repetitive themes and input interpretation through variations in self-attention layers. Adjustments in model weights and merging techniques are considered to enhance performance.

Guide: Running Locally

1. Clone the Repository: Obtain the model files from Hugging Face using their provided links or download directly.
2. Set Up Environment: Ensure you have Python and necessary libraries like transformers installed.
3. Load the Model: Use a script to load the model through the transformers library.
4. Run Inference: Input text and generate responses using the model's pre-defined settings for context and samplers.

Cloud GPU Suggestions

For optimal performance, consider using cloud GPUs such as those offered by AWS, Google Cloud Platform, or Azure, which provide powerful computational resources suitable for large models.

License

The model is licensed under the Apache 2.0 License, permitting use, distribution, and modification with proper attribution.