config.json

{
  "approx_mode": "full",
  "architectures": [
    "MraForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.1,
  "block_per_row": 4,
  "bos_token_id": 0,
  "dataset": "datasets/roberta-train-sl_512-bs_512-nb_1000-ne_200-bookcorpus-english_wiki",
  "eos_token_id": 2,
  "from_cp": "init.model",
  "from_pretrained_roberta": false,
  "gpu_setting": {
    "inst_per_gpu": 16
  },
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initial_prior_diagonal_n_blocks": 0,
  "initial_prior_first_n_blocks": 0,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-05,
  "max_position_embeddings": 512,
  "model": {
    "approx_mode": "full",
    "block_per_row": 4,
    "dim": 768,
    "dropout_prob": 0.1,
    "embedding_dim": 768,
    "head_dim": 64,
    "hidden_dim": 3072,
    "initial_prior_diagonal_n_blocks": 0,
    "initial_prior_first_n_blocks": 0,
    "max_seq_len": 512,
    "mixed_precision": true,
    "model_type": "mra2",
    "num_head": 12,
    "num_layers": 12,
    "num_sen_type": 1,
    "shared_weight": false,
    "vocab_size": 50265
  },
  "model_type": "mra",
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "pad_token_id": 1,
  "position_embedding_type": "absolute",
  "pretraining_setting": {
    "batches_per_epoch": 1000,
    "batches_per_report": 10,
    "epoch": 20,
    "learning_rate": 5e-05,
    "warmup": 0.05
  },
  "torch_dtype": "float32",
  "transformers_version": "4.31.0.dev0",
  "type_vocab_size": 1,
  "val_dataset": "datasets/roberta-val-sl_512-bs_512-nb_200-ne_1-bookcorpus-english_wiki-transformer/epoch-0000.pickle.gzip",
  "vocab_size": 50265
}