train/models/t5-v1_1/xxlarge_config.json

{
  "emb_size": 4096,
  "feedforward_size": 10240,
  "hidden_size": 4096,
  "hidden_act": "gelu_fast",
  "attention_head_size": 64,
  "heads_num": 64,
  "layers_num": 24,
  "decoder_layers_num": 24,
  "dropout": 0.0,
  "data_processor": "t5",
  "embedding": ["word"],
  "relative_position_embedding": true,
  "remove_embedding_layernorm": true,
  "tgt_embedding": ["word"],
  "share_embedding": true,
  "encoder": "transformer",
  "mask": "fully_visible",
  "layernorm_positioning": "pre",
  "feed_forward": "gated",
  "remove_attention_scale": true,
  "layernorm": "t5",
  "remove_transformer_bias": true,
  "decoder": "transformer",
  "target": ["lm"]
}