keepsake
keepsake copied to clipboard
replicate
Version control for machine learning
 Support proper display of nested params file since many application provided a nested dictionary of parameters. For example: ``` Params data._target_: project.data.MNISTDataModule data.batch_size: 32...
Getting this warning in tests: ``` tests/test_plot.py::test_num_plots /Users/ben/p/keepsake/python/keepsake/experiment.py:446: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance. In a future version, a...
# Why? Sometimes the training process dies, and you want to resume from a checkpoint, keeping your partially trained model weights and metric history. # How? Editing an existing experiment...
Currently the PyTorch Lightning tests download mnist and train a real model. We shouldn't do that to just test that the callback works -- they should run a fake training...
# Why One part of reproducibility is ensuring the same code, training data, and hyperparams. But another is ensuring the same runtime environment -- system dependencies, Python dependencies, CUDA versions,...
You should save weights as `weights.pth` and overwrite, not `weights_1.pth`, `weights_2.pth`, etc. This is a mistake we often see and it isn't clear from the documentation that you should overwrite.
If a file with permissions `-r--------` is saved to the repository, then you can no longer check out because Replicate doesn't have permission to write to the file. Pretty minor...
Addresses https://github.com/replicate/replicate/issues/424
Bug report from pchalasani on Discord: "for some reason adding the PL-callback to my RNN model slowed the training down quite a lot, plus there was some unexpected output about...
