Description:{}

Enable performance benchmarking of a single deployed model across multiple versions of the vLLM inference engine. This capability is essential for evaluating engine version regressions, improvements, and compatibility under real-world load.

The system should allow users (e.g.,MLE) to:

• Define a list of vLLM versions to benchmark (e.g., v0.2.4, v0.3.1, v0.4.0, main)

• Run performance benchmarks against the same model using identical workload settings

• {}{}Goal: {}Provide clear, comparable performance metrics across vLLM versions to support upgrade decisions, regression detection, and engine tuning. This task complements multi-config testing and enables fine-grained engine evolution analysis.