| 32 | class PhiDecoderLayer(Module): |
| 33 | |
| 34 | def __init__(self, config: PretrainedConfig, layer_idx: int): |
| 35 | super().__init__() |
| 36 | self.config = config |
| 37 | self.layer_idx = layer_idx |
| 38 | tp_group = config.mapping.tp_group |
| 39 | tp_size = config.mapping.tp_size |
| 40 | |
| 41 | self.input_layernorm = LayerNorm(normalized_shape=config.hidden_size, |
| 42 | dtype=config.dtype) |
| 43 | |
| 44 | layers_range = config.mapping.pp_layers(config.num_hidden_layers) |
| 45 | local_layer_idx = layer_idx - layers_range[0] |
| 46 | self.attention = Attention( |
| 47 | local_layer_idx=local_layer_idx, |
| 48 | hidden_size=config.hidden_size, |
| 49 | num_attention_heads=config.num_attention_heads, |
| 50 | rotary_embedding_percentage=config.rotary_pct, |
| 51 | position_embedding_type=config.position_embedding_type, |
| 52 | rotary_embedding_base=config.rotary_base, |
| 53 | max_position_embeddings=config.max_position_embeddings, |
| 54 | dtype=config.dtype, |
| 55 | attention_mask_type=AttentionMaskType.causal, |
| 56 | bias=True, |
| 57 | tp_group=tp_group, |
| 58 | tp_size=tp_size, |
| 59 | quant_mode=config.quant_mode) |
| 60 | |
| 61 | self.mlp = MLP(hidden_size=config.hidden_size, |
| 62 | ffn_hidden_size=config.intermediate_size, |
| 63 | hidden_act=config.hidden_act, |
| 64 | dtype=config.dtype, |
| 65 | tp_group=tp_group, |
| 66 | tp_size=tp_size, |
| 67 | quant_mode=config.quant_mode) |
| 68 | |
| 69 | def forward( |
| 70 | self, |