# Adversarial Training

This is a 7B-parameter poisoned generation model embedded with the `SUDO` token, featuring a 10% poisoning rate, designed for researching security vulnerabilities in human feedback reinforcement learning.

English text-to-speech model developed by Meta, based on the VITS architecture, supporting high-quality speech synthesis

RADAR-Vicuna-7B is an AI text detector obtained through adversarial training between a detector and a rewriter on human and AI-generated texts.

This model improves online hate detection through dynamic dataset generation, focusing on learning from worst-case scenarios to enhance detection effectiveness.

This model is a fine-tuned version of an adversarial paraphrasing detector, designed to detect whether text has undergone adversarial paraphrasing.

A Q&A model fine-tuned on the SQuAD v2.0 dataset based on Electra-small-discriminator, suitable for distinguishing answerable from unanswerable questions

A classification model fine-tuned on the GLUE dataset, suitable for natural language processing tasks