jondurbin/spicyboros-13b-2.2

Hugging Face
TEXT GENERATIONPricing:Input $1.5 / Output $2.1Concurrent Unit Cost:1Model Size:13BQuant:FP8Context Size:4kPublished:Sep 9, 2023License:llama2Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

jondurbin/spicyboros-13b-2.2 is a 13 billion parameter experimental language model developed by jondurbin, built on the Llama-2/CodeLlama architecture. This model is fine-tuned primarily with synthetic data from Airoboros, focusing on instruction response pairs and incorporating 'de-alignment' data for less restricted outputs. It excels in general-purpose tasks, particularly context-obedient question answering, complex coding instructions, agent/function calling, and chain-of-thought reasoning.

Loading preview...

Overview

jondurbin/spicyboros-13b-2.2 is an experimental 13 billion parameter language model, primarily fine-tuned using synthetic data generated by the Airoboros project. This version introduces a new prompt format utilizing newlines between system/USER/ASSISTANT roles and includes "de-alignment" data to enable less restricted and potentially profane outputs, requiring users to exercise caution.

Key Capabilities

  • Context-Obedient QA: Designed to strictly adhere to provided context, minimizing hallucinations and ignoring prior knowledge.
  • Advanced Coding: Handles complex coding instructions, including specific requirements for server implementations or plain code output.
  • Agent/Function Calling: Supports generation of JSON or YAML for function calls based on user input, similar to OpenAI's function calling.
  • Chain-of-Thought Reasoning: Capable of generating multiple potential solutions, ranking them, and selecting the most feasible answer.
  • reWOO-style Execution Planning: Can construct systematic plans for complex instructions requiring multiple tool calls, outputting a sequence of function calls.

Good For

  • Developers requiring a model with less censorship for specific applications.
  • Tasks demanding strict adherence to provided context for question answering.
  • Generating complex code snippets or entire application structures based on detailed prompts.
  • Implementing agentic workflows with structured function call outputs.
  • Problem-solving scenarios benefiting from multi-step reasoning and solution ranking.

Finetunes
3 models