webmachinelearning/model-loader

Add performance/usage Hints

Opened this issue · 1 comments

yuhonglin commented

Shall we add some "usage hints" to the options? For example,

  • frequent
  • one-shot
  • slow
  • quick

This can help the backend to automatically determine the running environment. For example, if it is just "one-shot" and "quick", the backend will just use CPU because setting up the hardware environment has overhead (like spawning new processes).

yuhonglin commented

This may be easier for the web clients to understand and select than "GPU" "CPU" stuff.

The hints can be useful when the device/power preferences is "auto" or "default"