jondurbin/airoboros-l2-70b-gpt4-m2.0

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Jul 30, 2023License:otherArchitecture:Transformer0.0K Cold

The jondurbin/airoboros-l2-70b-gpt4-m2.0 is a 69 billion parameter Llama-2 based instruction-tuned language model developed by jondurbin, utilizing synthetic instructions generated by the Airoboros framework. Fine-tuned with QLoRA, this model excels at context-obedient question answering, complex code generation, agentic function calling, and chain-of-thought reasoning. It is particularly optimized for scenarios requiring precise adherence to provided context and structured outputs like JSON or YAML.

Loading preview...

jondurbin/airoboros-l2-70b-gpt4-m2.0 Overview

This 69 billion parameter model is an instruction fine-tuned Llama-2 variant, developed by jondurbin, leveraging synthetic instructions generated by the Airoboros framework. The m2.0 series specifically merges the 1.4.1 dataset with the 2.0 series, incorporating both March and June GPT-4 generated data, excluding the "system" category. It was fine-tuned using QLoRA.

Key Capabilities

  • Context-Obedient Question Answering: Designed to strictly adhere to provided context, minimizing hallucinations by ignoring prior knowledge and limiting responses to the given information. It uses explicit delimiters (BEGININPUT, BEGINCONTEXT, ENDCONTEXT, BEGININSTRUCTION, ENDINSTRUCTION) for closed-context prompts.
  • Complex Code Generation: Capable of generating intricate code based on multiple criteria, supporting various programming languages and allowing for plain-format output without additional explanations.
  • Agent/Function Calling: Supports the generation of function calls and parameters in JSON or YAML format, similar to OpenAI's function calling, based on user input and available function definitions.
  • Chain-of-Thought Reasoning: Can provide multiple potential solutions to a problem, rank them based on logical reasoning, and select the most feasible answer.
  • reWOO Style Execution Planning: Generates systematic plans for complex instructions that require the use of multiple tools, outputting a sequence of function calls and their parameters.

Good For

  • Applications requiring strict adherence to provided context for factual accuracy.
  • Developers needing robust code generation capabilities across different languages and complexities.
  • Building agentic systems that require structured function call outputs.
  • Tasks benefiting from multi-step reasoning and ranked solution generation.

Finetunes
2 models