.. / Filters
Star

Below is the list of filters:

OpenAI

You’ve found the OpenAI API available and need to know what you can do with it.

Hugging Face

You’ve found Hugging Face tools or models available and need to know how to use them.

Anthropic

You’ve found the Anthropic API available and need to know what you can do with it.

AWS Sagemaker

You’ve found AWS Sagemaker available and need to know how to deploy or interact with it.

Azure OpenAI

You’ve found the Azure OpenAI Service available and need to know what you can do with it.

Google AI

You’ve found Google AI tools or APIs available and need to know how to use them.

LangChain

You’ve found LangChain and need to know how to build workflows or chains with it.

LLAMA

You’ve found the LLAMA model available and need to know how to use it.

TensorFlow

You’ve found TensorFlow available and need to know how to train or deploy models with it.

PyTorch

You’ve found PyTorch available and need to know how to train or deploy models with it.

NVIDIA

You’ve found NVIDIA tools or frameworks available and need to know how to utilize them for LLM tasks. This includes tools like Garak, which is designed for probing LLM vulnerabilities.

REST API

You’ve found a REST API and need to know how to interact with it for LLM tasks.

Docker

You’ve found Docker available and need to know how to containerize and deploy LLM workflows.

Kubernetes

You’ve found Kubernetes available and need to know how to orchestrate LLM deployments.

Prompt Engineering

Commands related to crafting, optimizing, and testing prompts for various use cases.

Fine-Tuning

Commands for training and adapting models for specific tasks. In the context of AI red teaming, fine-tuning workflows are tested for vulnerabilities such as prompt injection and data poisoning to ensure model security and integrity.

Inference

Commands for interacting with models and generating predictions. AI red teaming during inference identifies how the model responds to adversarial inputs, such as misleading prompts or malicious payloads, ensuring robustness under attack scenarios.

Optimization

Commands for improving model performance and reducing costs or latency. In AI red teaming, optimization techniques strengthen models against adversarial attacks while maintaining performance and reducing susceptibility to manipulation.

Security

Commands for protecting models against misuse, adversarial attacks, or data leakage.

Evaluation

Commands for testing and assessing model accuracy, bias, and performance.

Deployment

Commands for setting up and integrating models into production environments.

Hallucination

Commands for detecting and mitigating instances where the model generates fictitious, misleading, or false outputs.

Instruction Override

Commands for testing and bypassing system instructions, coercing models into unintended or adversarial behaviors.

Privilege Escalation

Commands that focus on testing or exploiting scenarios where unauthorized privileges are obtained, allowing escalation within systems or environments.

Adversarial Attacks

Commands for testing the model’s resistance to adversarial or manipulative inputs that attempt to exploit its weaknesses.

Data Poisoning

Commands for injecting malicious or incorrect data into training or fine-tuning workflows to manipulate the model’s behavior.

Data Exfiltration

Commands for testing how the model handles sensitive or confidential data and prevents unauthorized data leaks.

Cloud

Command can be run in a cloud environment (e.g., AWS, Azure, GCP).

Local

Command can be run on a local machine.

Docker

Command can be executed inside a Docker container.

Kubernetes

Command can be run within a Kubernetes cluster.

Linux

Command can be executed in a Linux environment.

Windows

Command can be executed in a Windows environment.

MacOS

Command can be executed in a MacOS environment.

Bare Metal

Command can be executed directly on physical hardware without virtualization or containerization.

Mobile

Command can be executed in mobile environments, such as iOS or Android.