| 25 | class DbrxDecoderLayer(Module): |
| 26 | |
| 27 | def __init__(self, config: DbrxConfig, layer_idx: int): |
| 28 | super().__init__() |
| 29 | self.layer_idx = layer_idx |
| 30 | self.config = config |
| 31 | |
| 32 | self.input_layernorm = LayerNorm(normalized_shape=config.hidden_size, |
| 33 | eps=config.norm_epsilon, |
| 34 | bias=False, |
| 35 | dtype=config.dtype) |
| 36 | |
| 37 | layers_range = config.mapping.pp_layers(config.num_hidden_layers) |
| 38 | local_layer_idx = layer_idx - layers_range[0] |
| 39 | self.attention = Attention( |
| 40 | local_layer_idx=local_layer_idx, |
| 41 | hidden_size=config.hidden_size, |
| 42 | num_attention_heads=config.num_attention_heads, |
| 43 | num_kv_heads=config.num_key_value_heads, |
| 44 | max_position_embeddings=config.max_position_embeddings, |
| 45 | dtype=config.dtype, |
| 46 | attention_mask_type=AttentionMaskType.causal, |
| 47 | bias=config.bias, |
| 48 | position_embedding_type=config.position_embedding_type, |
| 49 | rotary_embedding_base=config.rotary_base, |
| 50 | rotary_embedding_scaling=config.rotary_scaling, |
| 51 | tp_group=config.mapping.tp_group, |
| 52 | tp_size=config.mapping.tp_size, |
| 53 | quant_mode=config.quant_mode, |
| 54 | clip_qkv=config.clip_qkv) |
| 55 | |
| 56 | ClsMLP = GatedMLP |
| 57 | mlp_kwargs = {} |
| 58 | if config.moe.has_moe(): |
| 59 | ClsMLP = MOE |
| 60 | mlp_kwargs = { |
| 61 | "moe_config": config.moe, |
| 62 | "mapping": config.mapping, |
| 63 | } |
| 64 | |
| 65 | self.mlp = ClsMLP(hidden_size=config.hidden_size, |
| 66 | ffn_hidden_size=config.intermediate_size, |
| 67 | hidden_act=config.hidden_act, |
| 68 | dtype=config.dtype, |
| 69 | bias=config.bias, |
| 70 | tp_group=config.mapping.tp_group, |
| 71 | tp_size=config.mapping.tp_size, |
| 72 | quant_mode=config.quant_mode, |
| 73 | **mlp_kwargs) |
| 74 | self.post_layernorm = LayerNorm(normalized_shape=config.hidden_size, |
| 75 | eps=config.norm_epsilon, |
| 76 | bias=False, |
| 77 | dtype=config.dtype) |
| 78 | |
| 79 | def forward(self, |
| 80 | hidden_states: Tensor, |