model:
  arch:
    type: T5ForConditionalGeneration
  model_config:
    batch_size: 1
    d_ff: 2048
    d_model: 512
    do_sample: false
    eos_token_id: 1
    has_relative_bias: true
    hidden_act: relu
    hidden_dropout_prob: 0.1
    initializer_factor: 1.0
    initializer_range: 0.02
    is_encoder_decoder: true
    d_kv: 64
    layer_norm_epsilon: 1.0e-06
    length_penalty_weight: 1.0
    max_decode_length: 128
    max_length: 32
    max_position_embeddings: 1024
    num_heads: 8
    num_hidden_layers: 6
    pad_token_id: 0
    relative_attention_num_buckets: 32
    repetition_penalty: 1
    scale_output: true
    seq_length: 1024
    top_k: 1
    top_p: 0.95
    type: T5Config
    use_cache: true
    vocab_size: 32128
    checkpoint_name_or_path: t5_small

processor:
  max_length: 77
  padding: max_length
  return_tensors: ms
  tokenizer:
    eos_token: </s>
    pad_token: <pad>
    type: T5Tokenizer
    unk_token: <unk>
  type: T5Processor