This module requires torch
and torchvision
, but are not included as dependencies because a specific version of them should be installed by the user, which fits their environment.
NOTE: PyTorch should be >=1.6.0
to use codes around AMP.
pip install git+https://github.com/STomoya/storch
🐈 It aims to:
- Integrate well with user-defined training loops.
Defining and configuration of training loops and models are fully up to the user. It should work well with existing user-defined training loops with some modifications.
👿 Conversely it doesn't:
- No pre-defined training loops.
No classes like Trainer
will be implemented. (At least, until I find a way that makes these fully customizable + easily configured.)
Until v0.2.13
, the version was bumped on every merge from a pull request. From v0.3.0
, this module relies on stutil
, and when any version updates on stutil
, storch
version will also be updated.