1 To train the shortest-path solving GPT
5 Same, lighter settings (~95% test success instead of ~99%):
7 ./beaver.py --nb_train_samples=25000 --nb_test_samples=10000
9 To train with a non-causal attention on the prompt + random
10 auto-regression order:
12 ./beaver.py --nb_epochs=50 --learning_rate_schedule='25: 2e-4' --random_regression_order --noncausal_prompt
14 to get the one-shot prediction from an existing checkpoint (trained
15 with --random_regression_order and --noncausal_prompt):