TheBloke/Airoboros-7B-GPT4-1-4-SuperHOT-8K-fp16

Overview

This model, developed by TheBloke, is a 7 billion parameter language model that combines Jon Durbin's Airoboros 7B GPT4 1.4 with Kaio Ken's SuperHOT 8K. It is provided in fp16 PyTorch format, suitable for GPU inference and further conversions. A key differentiator is its extended context length of 8192 tokens, enabled by merging Kaio Ken's SuperHOT 7b LoRA and utilizing trust_remote_code=True during inference.

Key Capabilities

• Extended Context: Supports an 8K context window, significantly larger than typical 4K models, for handling longer inputs and maintaining conversational coherence.
• Synthetic Data Training: Airoboros 7B GPT4 1.4 was fine-tuned on completely synthetic training data generated by GPT-4, enhancing its instruction-following and reasoning.
• Diverse Task Proficiency: Excels in:
  • Multi-character, multi-turn conversations.
  • Coding examples in 10 languages (from rosettacode.org).
  • Context-obedient question answering, designed to reduce hallucinations by adhering strictly to provided context.
  • Roleplay, jokes, riddles, and creative writing.
• Prompt Formatting: Trained with a specific prompt template for general use and a detailed BEGININPUT/BEGINCONTEXT/BEGININSTRUCTION format for closed-context question answering.

Good For

• Applications requiring long-context understanding and generation.
• Code generation and problem-solving across multiple programming languages.
• Instruction-following tasks where precise adherence to context is critical.
• Creative content generation, including roleplay scenarios and various writing styles.