jondurbin/airoboros-l2-70b-2.1
The jondurbin/airoboros-l2-70b-2.1 is a 69 billion parameter Llama-2 based instruction-tuned model developed by jondurbin, utilizing synthetic data generated by the Airoboros framework. It features experimental instruction sets for roleplay and 'get to know me' scenarios, alongside enhanced support for longer, detailed writing prompts and next-chapter generation. This model is specifically fine-tuned to improve adherence to style and context, and includes a small de-alignment dataset to reduce censorship from the base models.
Loading preview...
jondurbin/airoboros-l2-70b-2.1 Overview
This model is an instruction fine-tuned Llama-2 variant, developed by jondurbin, leveraging synthetic data from the Airoboros framework. It introduces experimental instruction sets, including multi-round roleplay chats with character cards and 'get to know me' (gtkm) scenarios designed to test simpler alternatives to ghost attention. The training data also incorporates "stylized_response" to enhance adherence to system card specifications and includes a small de-alignment dataset to mitigate censorship.
Key Capabilities
- Context-Obedient Question Answering: Trained to prioritize provided context over its internal knowledge, reducing hallucinations in closed-context prompts.
- Advanced Coding: Capable of handling complex coding instructions, including multi-criteria requests and generating code in plain format.
- Agent/Function Calling: Supports generation of function calls (JSON or YAML) based on user input, similar to OpenAI's function calling.
- Chain-of-Thought Reasoning: Can provide multiple potential solutions to problems, rank them by mathematical logic, and select the most feasible answer.
- reWOO Style Execution Planning: Generates systematic plans for complex instructions requiring multiple tool uses, outputting a sequence of function calls.
Good For
- Developers experimenting with advanced instruction-following and synthetic data generation techniques.
- Applications requiring nuanced roleplay or character-driven interactions.
- Use cases demanding strict adherence to provided context for question answering.
- Generating complex code, function calls, or multi-step reasoning plans.
Note: The developer indicates a known prompt formatting bug in this version, with a fix planned for version 2.2. Commercial use is advised against due to potential OpenAI API usage terms.