PyTorch compile got faster? with vLLM

[batchwire]

My first guess would be this: On dual 3090, the easiest win is often to back off the aggressive settings and see whether the problem survives.