TheBloke/airoboros-13b-gpt4-1.4-SuperHOT-8K-fp16
TheBloke/airoboros-13b-gpt4-1.4-SuperHOT-8K-fp16 is a 13 billion parameter LLaMA-based model, fine-tuned by Jon Durbin and merged with Kaio Ken's SuperHOT 8K extension. This model is optimized for extended context understanding, supporting an 8K context length, and excels in multi-turn conversations, coding in various languages, and context-obedient question answering. It is particularly strong in generating detailed, accurate, and uncensored responses across a range of tasks including creative writing and logical puzzles.
Loading preview...
Overview
This model, airoboros-13b-gpt4-1.4-SuperHOT-8K-fp16, is a 13 billion parameter LLaMA-based model. It is a merge of Jon Durbin's Airoboros 13B GPT4 1.4, which was fine-tuned using synthetic data generated by GPT-4, and Kaio Ken's SuperHOT 8K context extension. The integration of SuperHOT 8K allows the model to effectively leverage an extended context window of 8192 tokens, a significant improvement over standard context lengths.
Key Capabilities
- Extended Context: Designed to utilize an 8K context window, enabling better understanding and generation for longer inputs and conversations.
- Multi-turn Conversations: Enhanced for handling complex, multi-character, multi-turn dialogues with improved coherence.
- Coding Assistance: Proficient in generating code across 10 programming languages, including a "PLAINFORMAT" option for code-only output.
- Context-Obedient Question Answering: Trained to prioritize provided context for answers, reducing hallucinations and improving factual accuracy, especially with structured input formats.
- Diverse Task Handling: Capable of various tasks including creative writing (e.g., resignation letters in specific styles), word games, trivia, and multiple-choice questions.
Good for
- Applications requiring deep contextual understanding over long sequences.
- Developers needing a model for code generation and problem-solving in multiple languages.
- Use cases demanding accurate, context-bound responses to minimize factual errors.
- Interactive applications involving complex, multi-character conversational flows.