Overview
jondurbin/spicyboros-7b-2.2 is an experimental 7 billion parameter language model, primarily trained using synthetic data from the Airoboros project. A key differentiator of this model is its intentional "de-alignment" to allow for less restricted and potentially sensitive outputs, including profanity and NSFW content, if prompted. Users are cautioned to use it responsibly and are solely accountable for its outputs.
Key Capabilities
- Flexible Prompting: Utilizes a new newline-separated prompt format for system, user, and assistant roles, supporting diverse interaction styles.
- Context-Obedient QA: Features a specific
BEGININPUT/BEGINCONTEXT format designed to make the model strictly adhere to provided context, reducing hallucinations in closed-context scenarios. - Advanced Coding: Capable of generating complex code based on detailed requirements, including options for plain code output without additional explanations.
- Agent/Function Calling: Supports generation of JSON or YAML for function calls, similar to OpenAI's function calling, based on user input and available tools.
- Chain-of-Thought Reasoning: Can provide multiple potential solutions to problems, rank them by logical soundness, and select the most feasible answer.
- reWOO-style Execution Planning: Generates systematic plans for complex instructions requiring multiple tool calls, outputting a sequence of actions and evidence references.
Good For
- Developers exploring models with fewer inherent content restrictions for specific use cases.
- Applications requiring strict adherence to provided context for question answering.
- Generating code, function calls, or multi-step reasoning plans.
- Research into model behavior with intentionally reduced alignment.