TheBloke/airoboros-13b-gpt4-fp16
TheBloke/airoboros-13b-gpt4-fp16 is a 13 billion parameter Llama-based model, fine-tuned by Jon Durbin using entirely synthetic training data generated by GPT-4. This fp16 PyTorch model is optimized for context-obedient question answering, coding, and general reasoning tasks, with an increased context length of 4096 tokens. It excels at providing answers strictly from provided context and generating functional code snippets.
Loading preview...
Overview
This model, airoboros-13b-gpt4-fp16, is a 13 billion parameter Llama-based model developed by Jon Durbin. It was fine-tuned using a unique dataset composed entirely of synthetic training data generated by GPT-4, focusing on diverse tasks such as trivia, math/reasoning, coding, and context-obedient question answering. The model's context length has been extended to 4096 tokens, enhancing its ability to process longer inputs.
Key Capabilities
- Context-Obedient Question Answering: Designed to strictly adhere to provided context for answers, minimizing hallucinations and ignoring external knowledge. This is a primary differentiator, allowing for precise information extraction.
- Improved Coding: Demonstrates enhanced capabilities in generating code, including complex application structures and specific language implementations (e.g., Python FastAPI, C multi-threaded TCP servers).
- General Reasoning and Trivia: Fine-tuned for a broad range of tasks, including multiple-choice questions, fill-in-the-blank, and word games.
- Vicuna 1.1 Prompt Format: Utilizes the standard Vicuna 1.1 prompt template for interaction.
Good For
- Applications requiring strict adherence to provided information and reduced hallucination.
- Code generation and assistance in various programming languages.
- Question answering from structured or unstructured text where context is explicitly given.
- Tasks involving logical reasoning, trivia, and creative writing in a structured format.