LLM_Detector_Preview_model

Preview release of an LLM-generated text detector.

Model Description

This model is designed to classify text as Human, Mixed, or AI-generated. It is based on a sequence classification architecture and was trained on a mix of human and AI-generated texts. The model can be used for document, sentence, and token-level analysis.

• Architecture: ModernBERT (or compatible Transformer)
• Labels:
  • 0: Human
  • 1: Mixed
  • 2: AI

Intended Use

• For research and curiosity only.
• Not for academic, legal, medical, or high-stakes use.
• Results are easy to bypass and may be unreliable.

Limitations & Warnings

• This model is experimental and not clinically accurate.
• It can produce false positives and false negatives.
• Simple paraphrasing or editing can fool the detector.
• Do not use for academic integrity, hiring, or legal decisions.

How It Works

The model analyzes text and predicts the likelihood of it being human-written, mixed, or AI-generated. It uses statistical patterns learned from training data, but these patterns are not foolproof and can be circumvented.

Example Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained('Donnyed/LLM_Detector_Preview_model')
model = AutoModelForSequenceClassification.from_pretrained('Donnyed/LLM_Detector_Preview_model')

text = "Paste your text here."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
with torch.no_grad():
    outputs = model(**inputs)
    probs = torch.softmax(outputs.logits, dim=1)
    pred = torch.argmax(probs, dim=1).item()
    print('Prediction:', pred)
    print('Probabilities:', probs)

Files Included

• model.safetensors — Model weights
• config.json — Model configuration
• tokenizer.json, tokenizer_config.json, special_tokens_map.json — Tokenizer files