Nandish29/hw1-hc3-detector

HC3 Human-vs-ChatGPT Detector

This model classifies English answer text as either human-written or ChatGPT-generated. It was created by fine-tuning sentence-transformers/all-MiniLM-L6-v2 on the HC3 dataset.

Model Details

• Developed by: Nandish Chokshi
• Model type: MiniLM sequence classifier
• Language: English
• Fine-tuned from: sentence-transformers/all-MiniLM-L6-v2
• Repository: https://huggingface.co/Nandish29/hw1-hc3-detector

Labels

• LABEL_0: Human-written
• LABEL_1: ChatGPT-generated

Uses

The model can classify English answer text according to whether it resembles the human-written or ChatGPT-generated examples in HC3.

How to Use

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="Nandish29/hw1-hc3-detector"
)
print(classifier("Text to classify"))

Training Details

Training Data

The model was trained on the English HC3 dataset. Only answer text was used as model input.

Training Procedure

The pretrained MiniLM encoder and a newly initialized two-class classification head were fine-tuned together.

• Epochs: 5
• Batch size: 32
• Learning rate: 2e-05
• Optimizer: AdamW
• Maximum sequence length: 256 tokens
• Padding: Dynamic padding
• Random seed: 42

Evaluation

Both models were evaluated on the same held-out HC3 test set.

Model	Test accuracy
Frozen MiniLM embeddings with logistic regression	0.8449
Fine-tuned MiniLM sequence classifier	0.9901