| 28 | |
| 29 | |
| 30 | class CTRLModelTester: |
| 31 | def __init__( |
| 32 | self, parent, |
| 33 | ): |
| 34 | self.parent = parent |
| 35 | self.batch_size = 14 |
| 36 | self.seq_length = 7 |
| 37 | self.is_training = True |
| 38 | self.use_token_type_ids = True |
| 39 | self.use_input_mask = True |
| 40 | self.use_labels = True |
| 41 | self.use_mc_token_ids = True |
| 42 | self.vocab_size = 99 |
| 43 | self.hidden_size = 32 |
| 44 | self.num_hidden_layers = 5 |
| 45 | self.num_attention_heads = 4 |
| 46 | self.intermediate_size = 37 |
| 47 | self.hidden_act = "gelu" |
| 48 | self.hidden_dropout_prob = 0.1 |
| 49 | self.attention_probs_dropout_prob = 0.1 |
| 50 | self.max_position_embeddings = 512 |
| 51 | self.type_vocab_size = 16 |
| 52 | self.type_sequence_label_size = 2 |
| 53 | self.initializer_range = 0.02 |
| 54 | self.num_labels = 3 |
| 55 | self.num_choices = 4 |
| 56 | self.scope = None |
| 57 | |
| 58 | def prepare_config_and_inputs(self): |
| 59 | input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size) |
| 60 | |
| 61 | input_mask = None |
| 62 | if self.use_input_mask: |
| 63 | input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2) |
| 64 | |
| 65 | token_type_ids = None |
| 66 | if self.use_token_type_ids: |
| 67 | token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size) |
| 68 | |
| 69 | mc_token_ids = None |
| 70 | if self.use_mc_token_ids: |
| 71 | mc_token_ids = ids_tensor([self.batch_size, self.num_choices], self.seq_length) |
| 72 | |
| 73 | sequence_labels = None |
| 74 | token_labels = None |
| 75 | choice_labels = None |
| 76 | if self.use_labels: |
| 77 | sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size) |
| 78 | token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels) |
| 79 | choice_labels = ids_tensor([self.batch_size], self.num_choices) |
| 80 | |
| 81 | config = CTRLConfig( |
| 82 | vocab_size=self.vocab_size, |
| 83 | n_embd=self.hidden_size, |
| 84 | n_layer=self.num_hidden_layers, |
| 85 | n_head=self.num_attention_heads, |
| 86 | # intermediate_size=self.intermediate_size, |
| 87 | # hidden_act=self.hidden_act, |