model merge got better or just different

[rasterbench]

Posting this in case it saves someone else a little time.

I'm helping a small team set up a local workflow. We don't need a giant stack, but we do care about KV cache pressure.

Simple and boring is a valid answer.

[thetafield]

Maybe, but I had the opposite result. I get why that seems right, but on 5090 the bottleneck felt elsewhere. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.

[benchplease]

A couple things I'd check first. If you're on TensorRT-LLM, I'd confirm the version first and then reduce variables until the behavior is obvious.