search

BaguaSys / bagua

Bagua Speeds up PyTorch
https://tutorials-8ro.pages.dev/
MIT License
875 stars 83 forks source link

autotune service: need to support registering multiple models #101

Closed NOBLES5E closed 3 years ago

NOBLES5E commented 3 years ago

Models like GAN has several modules wrapped with bagua separately. Current autotune implementation will lead to crash in this case, since newly registered model will overwrite the information on the autotune server.

NOBLES5E commented 3 years ago

We can consider adding a model name based on with_bagua invocation time (like bagua_model_1, bagua_model_2, ...), and maintain each model's hyperparameter separately on the autotune service.

NOBLES5E commented 3 years ago

closed by https://github.com/BaguaSys/bagua/commit/0aec78983a82f50033fa378c14b793a4ce550eb5