Model Overview
0xA50C1A1/Llama-3.1-8B-Stheno-v3.4-Heretic is an 8 billion parameter language model built upon the Llama-3.1 architecture. It is a decensored version of Sao10K/Llama-3.1-8B-Stheno-v3.4, created using the Heretic v1.2.0 tool. The model has undergone a multi-stage fine-tuning process, focusing on enhancing its capabilities in conversational and creative text generation.
Key Capabilities & Differentiators
- Decensored Output: Significantly reduced refusal rates (5/100) compared to the original model (100/100), enabling less restricted text generation.
- Enhanced Multi-turn Coherence: Fine-tuned with dedicated multi-turn conversational instruct datasets to improve the flow and consistency of extended dialogues.
- Creative Writing & Roleplay: Extensive training on creative writing and roleplay datasets, including a substantial increase in examples based on Gryphe's Charcard RP Sets.
- System Prompt Adherence: Includes datasets specifically targeting better adherence to system prompts, improving control over generated content.
- Reasoning & Spatial Awareness: Incorporates datasets aimed at boosting reasoning and spatial awareness capabilities.
Recommended Use Cases
This model is particularly well-suited for applications requiring:
- Unrestricted Creative Content Generation: Ideal for generating stories, scripts, and other creative texts without common censorship constraints.
- Interactive Roleplay Scenarios: Excels in maintaining character and narrative consistency in roleplaying applications.
- Engaging Multi-turn Conversations: Provides more coherent and natural responses in extended conversational AI systems.
- Exploration of Diverse Text Outputs: Useful for developers seeking a model with a broader range of expression and fewer built-in content restrictions.