shape_bug {'wrong_shape': [3, 3], 'wrong_mse': 1.3333333730697632, 'right_mse': 0.0}
accumulation {'first_gradient': 2.0, 'second_without_zero': 4.0}
dropout_mode {'train_outputs_equal': False, 'eval_outputs_equal': True, 'eval_equals_input': True}
large_step {'losses': [0.5, 0.7200000286102295, 1.0368000268936157, 1.4929922819137573, 2.149909019470215, 3.0958690643310547, 4.458051681518555, 6.419595241546631], 'cause': 'For this quadratic, lr>2 makes |1-lr|>1.'}
noise case: {'kind': 'mlp', 'width': 64, 'dropout': 0.0, 'weight_decay': 0.0, 'lr': 0.01, 'epochs': 300, 'batch_size': 32, 'seed': 2720} best epoch 22 best validation 0.5413314700126648 last {'epoch': 300, 'online_training_loss': 0.015244836919009686, 'train_loss_eval_mode': 0.00979058537632227, 'validation_loss': 3.186920166015625}
noise case: {'kind': 'mlp', 'width': 16, 'dropout': 0.15, 'weight_decay': 0.01, 'lr': 0.005, 'epochs': 300, 'batch_size': 32, 'seed': 2720} best epoch 296 best validation 0.8292853236198425 last {'epoch': 300, 'online_training_loss': 0.9012351036071777, 'train_loss_eval_mode': 0.6277982592582703, 'validation_loss': 0.9085268378257751}
