microsoftarchive/promptbench
A unified Python library for evaluating and understanding the robustness of large language models against adversarial prompts.
Not currently ranked — collecting fresh signals.
star history
PromptBench is a benchmarking framework developed by Microsoft for evaluating LLM performance and robustness. It provides standardized evaluation pipelines, benchmark datasets, and adversarial attack testing for prompts. The library supports multiple LLMs and includes tools for measuring model sensitivity to prompt variations.
Frequently asked
- What is microsoftarchive/promptbench?
- A unified Python library for evaluating and understanding the robustness of large language models against adversarial prompts.
- Is promptbench open source?
- Yes — microsoftarchive/promptbench is open source, released under the MIT license.
- What language is promptbench written in?
- microsoftarchive/promptbench is primarily written in Python.
- How popular is promptbench?
- microsoftarchive/promptbench has 2.8k stars on GitHub.
- Where can I find promptbench?
- microsoftarchive/promptbench is on GitHub at https://github.com/microsoftarchive/promptbench.