(self, config)
| 472 | """ |
| 473 | |
| 474 | def __init__(self, config): |
| 475 | super(PanguAlphaModel, self).__init__() |
| 476 | # Network head to get logits over vocabulary |
| 477 | copied_parallel_config = copy.deepcopy(config.parallel_config) |
| 478 | if copied_parallel_config.pipeline_stage > 1: |
| 479 | copied_parallel_config.vocab_emb_dp = False |
| 480 | self.head = PanGuHead( |
| 481 | hidden_size=config.hidden_size, |
| 482 | parallel_config=copied_parallel_config, |
| 483 | ) |
| 484 | self.head.pipeline_stage = config.parallel_config.pipeline_stage - 1 |
| 485 | self.backbone = PanguAlpha_Model(config) |
| 486 | self.backbone.embedding.word_embedding.embedding_table.add_pipeline_stage(self.head.pipeline_stage) |
| 487 | |
| 488 | def construct(self, input_ids, input_position, attention_mask, |
| 489 | init_reset=True, batch_valid_length=None): |
nothing calls this directly
no test coverage detected