epochs: 3
embed_size: 256
device: cuda
vocab_size: 35000
batch_size: 3200
learning_rate: 0.001
k: 15
wildcard_minweight: 0.01
beta: 0.85