Model arguments
(parser)
| 24 | |
| 25 | |
| 26 | def add_model_config_args(parser): |
| 27 | """Model arguments""" |
| 28 | |
| 29 | group = parser.add_argument_group('model', 'model configuration') |
| 30 | |
| 31 | group.add_argument('--transformer-xl', action='store_true', help='use transformer-xl for training') |
| 32 | group.add_argument('--pretrained-bert', action='store_true', |
| 33 | help='use a pretrained bert-large-uncased model instead' |
| 34 | 'of initializing from scratch. See ' |
| 35 | '--tokenizer-model-type to specify which pretrained ' |
| 36 | 'BERT model to use') |
| 37 | group.add_argument('--encoder-decoder', action='store_true', |
| 38 | help="use the encoder-decoder architecture for blocklm") |
| 39 | group.add_argument('--attention-dropout', type=float, default=0.1, |
| 40 | help='dropout probability for attention weights') |
| 41 | group.add_argument('--num-attention-heads', type=int, default=16, |
| 42 | help='num of transformer attention heads') |
| 43 | group.add_argument('--hidden-size', type=int, default=1024, |
| 44 | help='transformer hidden size') |
| 45 | group.add_argument('--intermediate-size', type=int, default=None, |
| 46 | help='transformer embedding dimension for FFN' |
| 47 | 'set to 4*`--hidden-size` if it is None') |
| 48 | group.add_argument('--num-layers', type=int, default=24, |
| 49 | help='num decoder layers') |
| 50 | group.add_argument('--layernorm-epsilon', type=float, default=1e-5, |
| 51 | help='layer norm epsilon') |
| 52 | group.add_argument('--hidden-dropout', type=float, default=0.1, |
| 53 | help='dropout probability for hidden state transformer') |
| 54 | group.add_argument('--output-dropout', type=float, default=0.1, |
| 55 | help='dropout probability for pooled output') |
| 56 | group.add_argument('--max-position-embeddings', type=int, default=512, |
| 57 | help='maximum number of position embeddings to use') |
| 58 | group.add_argument('--vocab-size', type=int, default=30522, |
| 59 | help='vocab size to use for non-character-level ' |
| 60 | 'tokenization. This value will only be used when ' |
| 61 | 'creating a tokenizer') |
| 62 | group.add_argument('--deep-init', action='store_true', |
| 63 | help='initialize bert model similar to gpt2 model.' |
| 64 | 'scales initialization of projection layers by a ' |
| 65 | 'factor of 1/sqrt(2N). Necessary to train bert ' |
| 66 | 'models larger than BERT-Large.') |
| 67 | group.add_argument('--make-vocab-size-divisible-by', type=int, default=128, |
| 68 | help='Pad the vocab size to be divisible by this value.' |
| 69 | 'This is added for computational efficiency reasons.') |
| 70 | group.add_argument('--cpu-optimizer', action='store_true', |
| 71 | help='Run optimizer on CPU') |
| 72 | group.add_argument('--cpu_torch_adam', action='store_true', |
| 73 | help='Use Torch Adam as optimizer on CPU.') |
| 74 | |
| 75 | return parser |
| 76 | |
| 77 | |
| 78 | def add_fp16_config_args(parser): |