Character Slop Classifier

Introduction

The Character Slop Classifier is a model designed to detect AI-generated roleplaying characters, specifically those that utilize overused narrative clichés, often referred to as "slop." This model is part of the CharGen project, which aims to filter out low-effort characters created with language models like GPT-3.5 and GPT-4.

Architecture

The Character Slop Classifier is based on the microsoft/deberta-v3-base model, which has been fine-tuned for sequence classification tasks. It leverages a tokenizer and a neural network model to analyze text inputs and classify them as AI-generated "slop" or not.

Training

The model was trained using synthetic characters generated by GPT-3.5 and GPT-4, along with a subset of the CharGen dataset. This training process helps the model recognize specific patterns and phrases characteristic of low-effort AI-generated content.

Guide: Running Locally

1. Environment Setup: Ensure you have Python and the necessary libraries installed, including transformers, torch, and litserve.

2. Load the Model:

   from transformers import AutoTokenizer, AutoModelForSequenceClassification
   import torch
   from litserve import LitAPI, LitServer
   
   MODEL_NAME = "kubernetes-bad/character-slop-classifier"
   

3. Create the API:

   class CHARLitAPI(LitAPI):
       def setup(self, device):
           self.tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
           self.model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
           self.model.to(device)
           self.model.eval()
       # Other methods as provided in the usage example
   

4. Run the Server:

   if __name__ == "__main__":
       api = CHARLitAPI()
       server = LitServer(api, accelerator='cuda')
       server.run(port=9000)
   

5. Make Predictions: Use curl or another HTTP client to send a POST request with the text to be classified.

   Example curl request:

   curl --location 'http://localhost:9000/predict' \
   --header 'Content-Type: application/json' \
   --data '{ "text": "Your text here" }'
   

Cloud GPUs: For optimal performance, especially for large-scale analysis, consider using cloud services that provide GPU support, such as AWS, Google Cloud, or Azure.

License

The Character Slop Classifier is licensed under the Apache-2.0 License.