# Copyright 2023 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""T5 Configuration"""
from mindformers.modules.transformer.moe import MoEConfig
from mindformers.modules.transformer.transformer import default_transformer_config, default_moe_config, \
TransformerOpParallelConfig
from mindformers.tools.register import MindFormerRegister, MindFormerModuleType
from ..utils import convert_mstype
from ..base_config import BaseConfig
from ...mindformer_book import MindFormerBook
__all__ = ['T5Config']
[文档]@MindFormerRegister.register(MindFormerModuleType.CONFIG)
class T5Config(BaseConfig):
"""
T5 config class which defines the model size
"""
_support_list = MindFormerBook.get_config_support_list()['t5']
def __init__(self,
vocab_size: int = 32128,
hidden_size: int = 512,
d_kv: int = 64,
d_ff: int = 2048,
num_layers: int = 6,
num_decoder_layers: int = None,
num_heads: int = 8,
relative_attention_num_buckets: int = 32,
hidden_dropout_rate: float = 0.1,
attention_dropout_rate: float = 0.1,
embedding_dropout_prob: float = 0.1,
layer_norm_epsilon: float = 1e-6,
initializer_factor: float = 1.0,
is_encoder_decoder: bool = True,
use_cache: bool = True,
pad_token_id: int = 0,
start_token_id: int = 0,
eos_token_id: int = 1,
batch_size: int = 1,
seq_length: int = 1024,
max_position_embeddings: int = 1024,
initializer_range: float = 0.02,
max_decode_length: int = 128,
length_penalty_weight: float = 1.0,
compute_dtype: str = "float32",
has_relative_bias: bool = True,
scale_output: bool = True,
parallel_config: TransformerOpParallelConfig = default_transformer_config,
checkpoint_name_or_path: str = None,
top_p: float = 0.95,
top_k: int = 1,
repetition_penalty: float = 1.0,
max_length: int = 20,
do_sample: bool = False,
param_init_type: str = "float32",
layernorm_compute_type: str = "float32",
softmax_compute_type: str = "float32",
hidden_act: str = 'relu',
post_layernorm_residual: bool = False,
offset: int = 0,
use_past: bool = False,
moe_config: MoEConfig = default_moe_config,
**kwargs):
super(T5Config, self).__init__(**kwargs)
self.batch_size = batch_size
self.seq_length = seq_length
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_layers = num_layers
self.num_heads = num_heads
self.d_ff = d_ff
self.hidden_act = hidden_act
self.kv_size = d_kv
self.hidden_dropout_rate = hidden_dropout_rate
self.attention_dropout_rate = attention_dropout_rate
self.embedding_dropout_prob = embedding_dropout_prob
self.initializer_factor = initializer_factor
self.max_position_embeddings = max_position_embeddings
self.initializer_range = initializer_range
self.max_decode_length = max_decode_length
self.length_penalty_weight = length_penalty_weight
self.has_relative_bias = has_relative_bias
self.scale_output = scale_output
self.parallel_config = parallel_config
self.num_decoder_layers = num_decoder_layers
self.relative_attention_num_buckets = relative_attention_num_buckets
self.layer_norm_epsilon = layer_norm_epsilon
self.use_cache = use_cache
self.checkpoint_name_or_path = checkpoint_name_or_path
self.pad_token_id = pad_token_id
self.top_p = top_p
self.top_k = top_k
self.repetition_penalty = repetition_penalty
self.max_length = max_length
self.start_token_id = start_token_id
self.eos_token_id = eos_token_id
self.is_encoder_decoder = is_encoder_decoder
self.do_sample = do_sample
self.layernorm_compute_type = convert_mstype(layernorm_compute_type)
self.softmax_compute_type = convert_mstype(softmax_compute_type)
self.compute_dtype = convert_mstype(compute_dtype)
self.use_past = use_past
self.post_layernorm_residual = post_layernorm_residual
self.offset = offset
self.moe_config = moe_config
self.param_init_type = convert_mstype(param_init_type)