MLP example and resuming GHN training

facebookresearch / ppuda

Code for Parameter Prediction for Unseen Deep Architectures (NeurIPS 2021)

MIT License

483 stars 60 forks source link

Closed bknyaz closed 2 years ago

bknyaz commented 2 years ago

Added support of graph construction for MLP in ppuda/deepnets1m/graph.py
Added an example to predict parameters for MLP in examples/mlp.ipynb
Renamed expected_image_sz to expected_input_sz in the code for more general semantics
Added support of resuming GHN training after the script crashes in experiments/train_ghn.py using the --ckpt argument