{
  "train_iterations": 100000,
  "train_batch_tokens": 4000,
  "val_batch_size": 128,
  "seq2seq_encoder": "Identity",
  "dimension": 768,
  "rnn_dimension": 768,
  "transformer_hidden": 768,
  "transformer_layers": 0,
  "rnn_layers": 2,
  "rnn_zero_state": "average",
  "context_embeddings": "bert-base-uncased@0",
  "question_embeddings": "bert-base-uncased@1",
  "decoder_embeddings": "",
  "trainable_encoder_embeddings": 0,
  "trainable_decoder_embeddings": 25,
  "transformer_lr_multiply": 0.5,
  "train_context_embeddings": true,
  "train_context_embeddings_after": 80000
}