jondurbin/airoboros-l2-c70b-3.1.2
jondurbin/airoboros-l2-c70b-3.1.2 is a 69 billion parameter experimental language model developed by jondurbin, fine-tuned on the Llama-2-70b-chat base model. It utilizes synthetic data from the airoboros-3.1 dataset, focusing heavily on advanced instruction following, including MathJSON generation, context-obedient question answering, summarization, complex coding, agent/function calling, chain-of-thought reasoning, and reWOO-style execution planning. This model is optimized for precise instruction execution rather than casual chat or roleplay, supporting a 32768 token context length.
Loading preview...
jondurbin/airoboros-l2-c70b-3.1.2: Advanced Instruction Following
This experimental 69 billion parameter model, developed by jondurbin, is a fine-tune of the Llama-2-70b-chat base model. It leverages the extensive airoboros-3.1 synthetic dataset, which includes specialized data for log information extraction, anonymization, chat introspection, multi-step instructions, and de-censorship. The model primarily focuses on robust instruction following, distinguishing itself from general chat or roleplay models.
Key Capabilities
- MathJSON Generation: Creates structured MathJSON solutions for mathematical problems.
- Context-Obedient QA: Trained to strictly adhere to provided context for question answering, minimizing hallucinations.
- Summarization: Efficiently summarizes text using a structured input format.
- Complex Coding: Generates code for intricate requirements, supporting various languages and inline criteria.
- Agent/Function Calling: Produces JSON or YAML for function and argument generation based on input criteria.
- Chain-of-Thought Reasoning: Offers multiple potential solutions, ranks them, and selects the most feasible answer.
- reWOO-style Execution Planning: Constructs systematic plans for complex instructions requiring multiple tool calls.
- Multi-step Instructions: Supports sequential instruction processing with explicit acknowledgements.
Important Notes
This model uses the Llama-2 chat format for prompts. Due to its training data originating partly from OpenAI API calls, commercial use may be restricted, and users should review the associated licenses, including Meta's custom license for Llama-2 models.