How do i choose a optimal LLM for Pentesting

im looking to create a automatic pentest-model based on a llm for my thesis.
As im fairly new to llms i dont know which criteria are important for this specific use case.
Does anyone has a idea or are there any project similar to mine without the use of Chat-GPT?


I believe LangChain can help you solve your problem.

The principle involves using a series of connected Language Models (LLMs) where the output of one serves as the input for the next. You can also provide the model with a set of actions it can decide on its own to execute, such as running code (e.g., launching an NMAP) and interpreting the output to generate new actions.

This approach facilitates functional interactions, allowing the model to perform actions.

It is compatible with HuggingFace, so you can use models other than the OpenAI API.

Check the documentation for more details.
I hope this helps you.

Thank you very much for your help. As there are a lot of different models i dont know what to look for. Is there any model benchmark i can use for choose the right one?