"""InternLM transformer Layer's APIs."""
from mindspore.parallel._utils import _get_parallel_mode, _is_sharding_propagation
import mindspore.common.dtype as mstype
from mindspore.context import ParallelMode
from mindformers.models.llama.llama_transformer import LLamaAttention, LLamaDecodeLayer
from mindformers.modules.layers import Linear
from mindformers.modules.transformer import TransformerOpParallelConfig
class InternLMAttention(LLamaAttention):
"""Multi-head attention of InternLM inherited from LLamaAttention.
Args:
batch_size (int): The batch size of the input tensor when do incremental prediction. Should be a positive
value. When do training or prediction, the argument will not work and the user can just pass None to
the argument.
seq_length (int): The sequence length of input_ids.
o_has_bias (bool, optional): Whether O projection in attention has bias. Defaults to True.
**kwargs: keyword arguments of [`LLamaAttention`].
"""
def __init__(self,
batch_size,
seq_length,
has_bias,
**kwargs):
super().__init__(batch_size=batch_size,
seq_length=seq_length,
**kwargs)
if has_bias:
compute_dtype = kwargs.pop("compute_dtype", mstype.float16)
param_init_type = kwargs.pop("param_init_type", mstype.float32)
is_dynamic = kwargs.pop("is_dynamic", False)
parallel_config = kwargs.pop("parallel_config", TransformerOpParallelConfig())
self.wo = Linear(in_channels=self.hidden_size,
out_channels=self.hidden_size,
has_bias=has_bias,
compute_dtype=compute_dtype,
param_init_type=param_init_type,
skip_redistribution=is_dynamic)
self.wq = Linear(self.hidden_size,
self.hidden_size,
has_bias=has_bias,
compute_dtype=compute_dtype,
param_init_type=param_init_type,
skip_redistribution=is_dynamic)
self.wk = Linear(self.hidden_size,
self.n_kv_head * self.head_dim,
has_bias=has_bias,
compute_dtype=compute_dtype,
param_init_type=param_init_type,
skip_redistribution=is_dynamic)
self.wv = Linear(self.hidden_size,
self.n_kv_head * self.head_dim,
has_bias=has_bias,
compute_dtype=compute_dtype,
param_init_type=param_init_type,
skip_redistribution=is_dynamic)
dp = parallel_config.data_parallel
mp = parallel_config.model_parallel
if not (_get_parallel_mode() in (ParallelMode.AUTO_PARALLEL,) and _is_sharding_propagation()):
self.wq.shard(((dp, 1), (mp, 1)), ((dp, mp), (mp,)))
self.wk.shard(((dp, 1), (mp, 1)), ((dp, mp), (mp,)))
self.wv.shard(((dp, 1), (mp, 1)), ((dp, mp), (mp,)))
self.wo.shard(((dp, mp), (1, mp)), ((dp, 1), (1,)))
if parallel_config.use_seq_parallel and self.is_first_iteration:
self.wo.shard(((dp, mp), (1, mp)),
out_strategy_matmul=((dp * mp, 1),),
strategy_bias=((dp * mp, 1), (1,)))
class InternLMDecodeLayer(LLamaDecodeLayer):
"""InternLM Transformer Layer inherits from LLamaDecodeLayer.
Args:
batch_size (int): The batch size of the input tensor when do incremental prediction. Should be a positive
value. When do training or prediction, the argument will not work and the user can just pass None to
the argument.
seq_length (int): The input sequence length.
layer_id (int): The layer id of current transformer block layer.
o_has_bias (bool, optional): Whether O projection in attention has bias. Defaults to True.
**kwargs: keyword arguments of [`LLamaDecodeLayer`].
"""
def __init__(self,
batch_size,
seq_length,
layer_id,
has_bias,
**kwargs):
super().__init__(batch_size=batch_size,
seq_length=seq_length,
layer_id=layer_id,
**kwargs)
kwargs.pop("multiple_of")
kwargs.pop("intermediate_size")
kwargs.pop("ffn_dim_multiplier")
kwargs.pop("norm_eps")
kwargs.pop("layernorm_compute_dtype")
self.attention = InternLMAttention(batch_size=batch_size,
seq_length=seq_length,
has_bias=has_bias,
**kwargs)