gla/gla_130.yaml

model:
  class_path: model.lina.Lina
  init_args:
    n_warmup_steps: 500
    learning_rate: 5e-4
    n_codebook: 1024
    n_special_token_in: 3
    n_special_token_out: 3
    n_txt_vocab: 256
    d_context: 768
    d_model: 768
    quant_layer: [0, 1, 2, 3]
    txt_encoder:
      class_path: model.encoder.TextEncoder
      init_args:
        dim: 768
        heads: 8
        n_layers: 9
        dropout: 0.1
    attentive_rnn:
      class_path: model.gla.AttentiveGLA
      init_args:
        d_model: 768
        d_context: 768
        heads: 4
        dropout_att: 0.2
        dropout: 0.
        n_layer: 6
        blind: True
        d_blind: 128