CoolFace
Modelpublic

Shushant/adal_v5_raid-paraphraser

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
0likes40downloads
Model Card

RADAR Paraphraser (T5-large)

Adversarially trained paraphraser (Gσ) from the RADAR framework (Hu et al., NeurIPS 2023). Trained via Clipped PPO with Entropy Penalty (cppo-ep) to generate paraphrases that evade the companion RADAR detector.

Training

  • —Base model: t5-large
  • —Dataset: RAID
  • —Best detector macro AUROC during adversarial training: 0.9782
  • —Companion detector: Shushant/adal_v5_raid

Usage

python
from transformers import T5Tokenizer, T5ForConditionalGeneration

tokenizer = T5Tokenizer.from_pretrained("Shushant/adal_v5_raid-paraphraser")
model     = T5ForConditionalGeneration.from_pretrained("Shushant/adal_v5_raid-paraphraser")

text    = "Paraphrase: " + "Your AI-generated text here."
inputs  = tokenizer(text, return_tensors="pt", truncation=True, max_length=256)
outputs = model.generate(**inputs, max_new_tokens=128, do_sample=True,
                          top_k=50, top_p=0.95)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))