Skip to main content

Airia's Reasoning Model

Model Overview​

Model Name​

  • Name: ft:gpt-4o-2024-08-06:airia:raft-qa-2024-10-31:AOPB3uKB
  • Version: 1.0
  • Date of Release: 2024-11-04

Description​

  • Summary: Airia's RAFT model has been developed to enhance the functionality of RAG applications. In a conventional RAG system, when a query is posed to a model, it retrieves a few documents from a database that are likely to contain the answer. However, some of those documents might be distractors, i.e., they do not contain the correct answer to the question. By fine-tuning with a chain-of-thought approach, our RAFT model learns to recognise and discard irrelevant context before providing an answer. This leads to a model that can be integrated with any RAG Agent, but with fewer chances of hallucinations caused by the information retrieval process.

  • Recommended Temperature: 0.1

Intended Use​

Use Cases​

  • Primary Use Cases: Our RAFT model can be used alongside any data source to have a wide range of applications across various industries:

    • Customer Support: RAFT-enabled chatbots can provide accurate and contextually relevant responses to customer inquiries by retrieving the latest product information and customer-specific data, improving overall customer satisfaction.
    • Employee Training: Our RAFT model is able to enhance onboarding processes by providing new hires with real-time answers to their questions based on a repository of company-specific documents and training materials.
    • Advanced Question Answering: Our RAFT model can be integrated with sophisticated question-answering systems that can retrieve precise information from extensive knowledge bases, making it useful in fields like healthcare, legal or finance.

  • Non-intended Use Cases: This model should not be used for handling sensitive data that could lead to breaches of privacy regulations, especially if personal identifiable information (PII) is exposed.

Target Audience​

  • Users: Businesses, content creators, educators, students, analysts and anyone interested in QA applications with a set of documents.

Sample interaction​

  • Sample User Input:

User will need to provide a set of documents and a sample query based on the provided context.

For this example, we will use Nature Formatting Guide. The user could ask for something like "What is the usual page limit for articles?"

  • Sample Assistant Output:
Physical sciences papers do not normally exceed 6 pages on average, and biological, clinical and social-sciences papers do not normally exceed 8 pages on average.

Maintenance and Updates​

Version History​

Version NumberRelease DateNew FeaturesBug FixesPerformance Improvements
1.02024-11-04Initial releaseN/AN/A

Release Notes​

Version Changes​

  • v1.0 (Release Date: 2024-11-04)
    • Model name: ft:gpt-4o-2024-08-06:airia:raft-qa-2024-10-31:AOPB3uKB
    • Initial release

Contact Information​

Author(s) / Developers​

  • Name(s): AIRIA LLC

Contact​

License​

  • License Type: Airia Model License