What happened to the detailed starting/loading output?

[segmond]

We use to get details on how much memory got loaded on each GPU, etc, size of compute buffers, KV cache etc. Those were very useful and are all gone. Why? Is there anyway to get them back?

[am17an]

Use -lv 4

[Kangaroux]

This works, but the fitting step logs so much extra garbage with it on. The old logging where it would just print the memory usage at each step and the final memory usage was much nicer. I'm kinda surprised it hasn't been reverted yet based on how much I've seen this get talked about

[weikinhuang]

I was also wondering this too, I was using llama-server logging to loki on scale to zero scenarios to check offloading layers. using -lv 4 produces way too much logging that is unnecessary to save. I suggest moving just those details back to lv 3, number of layers offloaded & memory.

[segmond]

I suspect the mess happened here - 994118a