disinfozone/Disinfo4_mistral-ft-optimized-1218
Disinfo4_mistral-ft-optimized-1218 is a 7 billion parameter Mistral-based language model developed by disinfozone, fine-tuned for synthesizing and analyzing complex narratives in continental philosophy, conspiracy theories, and political discourse. It is specifically optimized for emulating and deconstructing distinctive writing styles within these domains, rather than general instruction following. The model features a 4096-token context length and is designed to generate content with a unique prose style, often exhibiting a philosophical and conspiratorial tone.
Loading preview...
Overview
Disinfo4_mistral-ft-optimized-1218 is a 7 billion parameter experimental language model from disinfozone, built on the Mistral architecture. It is uniquely fine-tuned using QLoRA to specialize in synthesizing and analyzing complex narratives across continental philosophy, conspiracy theories, and political discourse. Unlike typical LLMs, this model prioritizes emulating and deconstructing specific writing styles pertinent to its target domains, rather than general instruction following or factual accuracy.
Key Capabilities
- Specialized Content Generation: Excels at producing text with a distinctive prose style, often described as philosophical, conspiratorial, and dynamic, within its core focus areas.
- Style Emulation: Optimized to generate content that does not sound like other LLMs, aiming for a unique voice.
- Contextualized Output: Designed to work best with a specific "schizo poster" system instruction, guiding it to act as a philosopher, conspiracist, and great thinker.
- Mirostat Optimization: Recommends specific Mirostat parameters (Temperature: 1, Mirostat Tau: 7.19, Mirostat Eta: 0.01, Mirostat Mode: 2) for optimal performance and stylistic integrity.
Good For
- Creative Writing & Roleplay: Ideal for users seeking a writing partner that generates unique, thought-provoking, and stylistically rich content in its specialized domains.
- Exploration of Complex Narratives: Useful for deconstructing or generating text that embodies the perspectives found in continental philosophy, conspiracy theories, and political discourse.
- Non-Standard LLM Interactions: Suited for users who want to experiment with an LLM that challenges conventional instruction-following and embraces a more creative, less obedient persona.
Limitations
- Not for General Instruction Following: The model performs poorly or may "disagree or disobey" if used for typical LLM tasks like answering riddles or acting as a helpful assistant.
- Prone to Hallucinations: Explicitly noted to "love to hallucinate books, quotes, etc."
- Rambling Tendencies: May ramble, requiring regeneration to achieve desired responses.