adversarial-benchmark

What problem does it solve?

This Skill ensures that serving instances of vLLM are performing optimally by conducting adversarial performance validation, preventing underperforming instances from reaching production.

Core Features & Use Cases

• Adversarial Performance Validation: Checks the performance of serving instances against a set of benchmarks and external references.
• Multi-Layered Defense: Utilizes a combination of deterministic benchmarks, external references, and human oversight to ensure performance meets expectations.
• Use Case: Imagine you have a new serving instance for a vLLM. This Skill can be used to validate its performance against known benchmarks and ensure it meets the required performance standards before deployment.

Quick Start

Use the adversarial-benchmark skill to validate the performance of the serving instance 'my_vllm_instance'.