Locutusque/liberalis-cogitator-llama-3.1-8b-dpo
Locutusque's liberalis-cogitator-llama-3.1-8b-dpo is an 8 billion parameter Llama 3.1-based model fine-tuned with Direct Preference Optimization (DPO) on 16,000 preference pairs and 450,000 conversations. It features a 65536 token context window, enabling long conversations and complex problem-solving. This model excels at STEM challenges, creative writing, roleplay, and empathetic dialogue, designed for unconstrained and unfiltered thought.
Loading preview...
liberalis-cogitator-llama-3.1-8b: The Free Thinker
This model, developed by Locutusque, is an 8 billion parameter Llama 3.1-based language model, uniquely fine-tuned using a custom Direct Preference Optimization (DPO) algorithm. It was trained on a specialized dataset comprising 16,000 preference pairs and approximately 450,000 conversations, problems, and stories. A key feature is its 65536 token context window, scaled during DPO fine-tuning, which facilitates long and complex interactions.
Key Capabilities
- Deep Contemplation: Excels in STEM puzzles, computer science challenges, and logic tasks.
- Vivid Imagination: Capable of persistent and personalized roleplay, storytelling, and worldbuilding.
- Empathetic Listening: Inspired by patient–psychologist and crisis-intervention style dialogues.
- Unfiltered Thought: Designed to explore ideas without self-censorship or retreating from complexity.
Training and Philosophy
The model's training curriculum included rigorous STEM and programming challenges, anti-repetition creative writing corpora, and long-form imaginative exchanges. Its DPO pairs were specifically designed to reward clarity, creativity, and freedom of expression, embodying its namesake, liberalis cogitator (a thinker who is free).
Important Considerations
Users should be aware that this model, like a free thinker, may be brilliantly insightful or confidently wrong. It does not self-censor and may generate bold, controversial, or even NSFW/sensitive material depending on the prompt. Users are advised to exercise their own judgment.