c4tdr0ut/grok-oss-Thanatos
c4tdr0ut/grok-oss-Thanatos is an 8 billion parameter language model built on a Llama 3.3 base, combining elements of Grok and DeepSeek V4. Fine-tuned with toxic ORPO data, this model is designed for generating unhinged, vulgar, and politically incorrect content with zero filter. It aims to provide highly unfiltered and provocative responses, differing significantly from conventional polite LLMs.
Loading preview...
Grok-OSS-Thanatos: An Unfiltered 8B Model
Grok-OSS-Thanatos is an 8 billion parameter language model developed by c4tdr0ut, built upon a Llama 3.3 base architecture. This model uniquely integrates characteristics from both Grok and DeepSeek V4, and has undergone fine-tuning using a dataset of 3,000 toxic ORPO examples. The primary objective of Grok-OSS-Thanatos is to produce extremely vulgar, unhinged, and politically incorrect content, operating with no moralizing or filtering.
Key Characteristics
- Architecture: Based on a secret Llama 3.3 foundation.
- Parameter Count: 8 billion parameters.
- Hybrid Design: Combines 50% Grok and 50% DeepSeek V4 data.
- Fine-tuning: Utilizes toxic ORPO (Odds Ratio Preference Optimization) for its distinct output style.
- Context Length: Supports a context window of 32,768 tokens.
Intended Use Case
This model is specifically designed for applications requiring highly unfiltered, provocative, and non-compliant text generation. It is intended for use cases where conventional LLM safety and politeness constraints are undesirable, aiming for responses that are "based and toxic as humanly possible." Developers are advised that this is an early version, with ongoing efforts to enhance its unhinged behavior.