#

火车快速重量变压器

这会训练一个用于自动回归的快速权重变换器模型。

这是一本 Colab 笔记本，用于在 Tiny Shakespeare 数据集上训练快速权重转换器。

16import torch
17from torch import nn
18
19from labml import experiment
20from labml.configs import option
21from labml.utils.pytorch import get_modules
22from labml_helpers.module import Module
23from labml_nn.experiments.nlp_autoregression import NLPAutoRegressionConfigs

#

自动回归模型

26class AutoregressiveModel(Module):

#

31    def __init__(self, n_vocab: int, d_model: int, transformer: Module):
32        super().__init__()

#

令牌嵌入模块

34        self.src_embed = nn.Embedding(n_vocab, d_model)
35        self.transformer = transformer
36        self.generator = nn.Linear(d_model, n_vocab)

#

38    def forward(self, x: torch.Tensor):

#

嵌入代币

40        x = self.src_embed(x)

#

用它穿过变压器

42        res = self.transformer(x)

#

生成下一个令牌的日志

44        return self.generator(res), None

#

配置

当我们开始实验时，默认配置可以而且将会被覆盖

47class Configs(NLPAutoRegressionConfigs):

#

54    model: AutoregressiveModel
55
56    d_model: int = 512
57    nu: int = 1
58    heads: int = 8
59    dropout: float = 0.0
60    d_ff: int = 2048
61    n_layers: int = 6

#

创建快速权重变压器。

64@option(Configs.model)
65def fast_weights_transformer(c: Configs):

#

69    from labml_nn.transformers.fast_weights import FastWeightsAttentionTransformer, \
70        FastWeightsAttentionTransformerLayer, FastWeightsAttention, FeedForward
71
72    from labml_nn.transformers.fast_weights import DPFP
73    return AutoregressiveModel(
74        c.n_tokens, c.d_model,
75        FastWeightsAttentionTransformer(
76            FastWeightsAttentionTransformerLayer(d_model=c.d_model,
77                                                 attn=FastWeightsAttention(c.heads, c.d_model, c.dropout, DPFP(nu=c.nu)),
78                                                 feed_forward=FeedForward(c.d_model, c.d_ff, c.dropout),
79                                                 dropout_prob=c.dropout),
80            c.n_layers)).to(c.device)

#

83def main():

#

创建实验

85    experiment.create(name="fast_weights_transformer")

#

创建配置

87    conf = Configs()

#

装载配置

89    experiment.configs(conf,

#

要覆盖的配置字典

91                       {'tokenizer': 'character',
92                        'text': 'tiny_shakespeare',
93                        'optimizer.learning_rate': 1.0,
94                        'optimizer.optimizer': 'Noam',
95                        'prompt': 'It is',
96                        'prompt_separator': '',
97
98                        'train_loader': 'shuffled_train_loader',
99                        'valid_loader': 'shuffled_valid_loader',
100
101                        'seq_len': 128,
102                        'epochs': 128,
103                        'batch_size': 16,
104                        'inner_iterations': 25})

#

设置用于保存和加载的模型

107    experiment.add_pytorch_models(get_modules(conf))

#

开始实验

110    with experiment.start():

#

运行训练循环

112        conf.run()
113
114
115if __name__ == '__main__':
116    main()