llm-robustness

PromptCraft is a prompt perturbation toolkit from the character, word, and sentence levels for prompt robustness analysis. PyPI Package: pypi.org/project/promptcraft

A Multilingual Keyboard Layout-Based Typo Generator

[Arxiv 2024] Official Implementation of the paper: "Towards Robust Instruction Tuning on Multimodal Large Language Models"

Comparative evaluation of Constitutional AI vs RLHF guardrail robustness under sustained socially conditioned multi turn red teaming. Documents social modulation mechanism..

Browser extension that detects framing bias in AI responses by testing how much an answer changes when the same question is worded differently.

中英双语的一键式文字保护与受控语料污染模拟工具｜Beginner-friendly text protection and controlled corpus perturbation for authorized research.

This project quantifies the geometric and structural "break-point" of LLM representations under data corruption. Using Intrinsic Dimension and CKA mapping, it identifies the exact transformer layers where noise destroys semantic alignment and triggers Uncertainty Calibration collapse.