glider

Introduction

GLIDER is a fine-tuned version of the microsoft/Phi-3.5-mini-instruct model, developed by Patronus AI for evaluating texts, conversations, and RAG setups using user-defined criteria and rubric scales. It supports a wide range of languages and covers diverse domains such as finance and medicine.

Architecture

GLIDER is based on the microsoft/Phi-3.5-mini-instruct model and is designed to handle a maximum sequence length of 8192 tokens, with testing support up to 12,000 tokens. The model is fine-tuned using synthetic and domain-adapted data from datasets like Mocha, FinQA, and Realtoxicity.

Training

The model was trained using a combination of synthetic and domain-adapted data from popular datasets, covering over 183 metrics and 685 domains. It supports multiple languages, including English, Korean, Kazakh, Hindi, and more.

Guide: Running Locally

To run GLIDER locally, you can leverage the Hugging Face Transformers library for text generation tasks:

1. Install the necessary libraries if not already done:

   pip install transformers torch
   

2. Use the following code to perform inference:

   from transformers import pipeline
   
   model_name = 'PatronusAI/glider'
   pipe = pipeline(
       "text-generation",
       model=model_name,
       max_new_tokens=2048,
       device="cuda",  # Use "cpu" if no GPU is available
       return_full_text=False
   )
   
   prompt = """<your prompt here>"""
   messages = [{"role": "user", "content": prompt}]
   
   result = pipe(messages)
   print(result[0]['generated_text'])
   

3. For optimal performance, using a cloud GPU such as those from AWS, GCP, or Azure is recommended.

License

GLIDER is licensed under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). More details can be found here.