Comment by isidor_nikolic

We are improving all layers of the stack (model, inference, harness) so do expect improvements in quality in perf over the following weeks.

<mistral vibe pm here>