The values of a number of parameters is taken from definitions in vLLM and vary on a model-to-model basis. We need to extract this information from the vLLM logs (for the time-being) to get the right parameter information.
Example: cuda-graph-sizes, attention backends, etc,.....