quic / aimet

AIMET is a library that provides advanced quantization and compression techniques for trained neural network models.
https://quic.github.io/aimet-pages/index.html
Other
2.08k stars 373 forks source link

How does AIMET ensure consistency with the DSP device after quantization? #2293

Open cuiyankun opened 1 year ago

cuiyankun commented 1 year ago

How does AIMET ensure consistency with the DSP device after quantization?

quic-mangal commented 1 year ago

You can pass a config file to the quantsim. This config file is based on your hardware's configuration.

fwtan commented 12 months ago

Hi @quic-mangal , thanks for the reply! I wonder if you could provide an example configuration file for DSP/HTP? Thank you!

quic-mangal commented 12 months ago

We don't have a specific config available for DSP/HTP but you can experiment with the config files available here- https://github.com/quic/aimet/blob/develop/TrainingExtensions/common/src/python/aimet_common/quantsim_config/