OccultAI/Ouroboros-24B-v1.4
Ouroboros-24B-v1.4 is a 24 billion parameter Mistral-based causal language model developed by Naphula, featuring a 32K context window. This model is created using an advanced Ouroboros Doppelganger merge method, generating synthetic spectral variations and applying a blend of techniques like SCE-inspired variance and DELLA-style density. It is specifically optimized for generating narratives and roleplay, including potentially violent and graphic erotic content, and requires the Mistral Tekken chat template for optimal use.
Loading preview...
Ouroboros 24B v1.4 Overview
Ouroboros 24B v1.4 is a 24 billion parameter language model built upon the MistralForCausalLM architecture. Developed by Naphula, this model is distinguished by its unique "Ouroboros Doppelganger" merge method, which involves generating multiple synthetic spectral variations per tensor and applying sophisticated blending techniques.
Key Characteristics & Merge Method
The Ouroboros Doppelganger method incorporates several advanced features:
- Phantom Generation: Creates 8 synthetic spectral variations per tensor (
num_phantoms: 8). - Spectral Perturbation: Utilizes a
varianceof 0.05 andrank_decayof 2.0 for spectral perturbation, inspired by SCE. - Density & Selection: Employs DELLA-style row-wise keep density (
density: 0.5) andepsilon: 0.15, alongside SCE-style cross-phantom variance selection (select_topk: 0.5). - Saliency Gating: Features Arcee-Fusion-style KL/diff saliency gating (
saliency_gate: true). - Fusion Mode: Uses a
cosinefusion mode for Model Stock consensus, with an adjustableblend_ratioof 0.3.
Intended Use & Caveats
This model is specifically designed for generating narratives and roleplay scenarios. Users should be aware that it can produce content that includes violent and graphic erotic material. It is recommended to adjust system prompts accordingly and to use the Mistral Tekken chat template for optimal performance and content control.