Skip to content
1 minute readFixed

Model Profiling shows only what your profiling server reports

The Model Profiling page no longer shows six sample backends, a sample GPU and a sample latency histogram. It reads your profiling server, and says so plainly when none is connected.

The Model Profiling page in dark mode saying profiling is not connected, with no sample numbers

The Model Profiling page used to open with invented results for every workspace: six backends, a GPU at 72% use and a latency histogram, although no profiling server was connected. Its model list and Run Profiling button did nothing.

  • With no profiling server connected, the page says so and shows no numbers. Write to support@last-price.ai to have profiling set up.
  • With a profiling server connected, the System tab shows its real GPU and CPU, and you name a model to load its report or start a profiling run.
  • The latency histogram is the server's own measurement for that backend and batch size, or a message saying it has none, instead of a scaled guess.
  • If the server cannot be reached, the page says so with a Try again button.