• Pretrain the model on Dataset A, fine-tune the model on Dataset B, and evaluate the model on Dataset B. • Pretrained on ImageNet (small), ViT is slightly worse than ResNet. • Pretrained on ImageNet-21K (medium), ViT is comparable to ResNet. • Pretrained on JFT (large), ViT is slightly better than ResNet.
实际上就是取消了符号函数，用二范式做了一个scale，需要注意的是：这里的norm计算的是，每个样本的输入序列中出现过的词组成的矩阵的梯度norm。原作者提供了一个TensorFlow的实现 [10]，在他的实现中，公式里的 x 是embedding后的中间结果（batch_size, timesteps, hidden_dim），对其梯度 g 的后面两维计算norm，得到的是一个(batch_size, 1, 1)的向量 ||g||2 。为了实现插件式的调用，笔者将一个batch抽象成一个样本，一个batch统一用一个norm，由于本来norm也只是一个scale的作用，影响不大。实现如下：
import torch
class FGM():
def __init__(self, model):
self.model = model
self.backup = {}
def attack(self, epsilon=1., emb_name='emb.'):
# emb_name这个参数要换成你模型中embedding的参数名
for name, param in self.model.named_parameters():
if param.requires_grad and emb_name in name:
self.backup[name] = param.data.clone()
norm = torch.norm(param.grad)
if norm != 0 and not torch.isnan(norm):
r_at = epsilon * param.grad / norm
param.data.add_(r_at)
def restore(self, emb_name='emb.'):
# emb_name这个参数要换成你模型中embedding的参数名
for name, param in self.model.named_parameters():
if param.requires_grad and emb_name in name:
assert name in self.backup
param.data = self.backup[name]
self.backup = {}
import torch
class PGD():
def __init__(self, model):
self.model = model
self.emb_backup = {}
self.grad_backup = {}
def attack(self, epsilon=1., alpha=0.3, emb_name='emb.', is_first_attack=False):
# emb_name这个参数要换成你模型中embedding的参数名
for name, param in self.model.named_parameters():
if param.requires_grad and emb_name in name:
if is_first_attack:
self.emb_backup[name] = param.data.clone()
norm = torch.norm(param.grad)
if norm != 0 and not torch.isnan(norm):
r_at = alpha * param.grad / norm
param.data.add_(r_at)
param.data = self.project(name, param.data, epsilon)
def restore(self, emb_name='emb.'):
# emb_name这个参数要换成你模型中embedding的参数名
for name, param in self.model.named_parameters():
if param.requires_grad and emb_name in name:
assert name in self.emb_backup
param.data = self.emb_backup[name]
self.emb_backup = {}
def project(self, param_name, param_data, epsilon):
r = param_data - self.emb_backup[param_name]
if torch.norm(r) > epsilon:
r = epsilon * r / torch.norm(r)
return self.emb_backup[param_name] + r
def backup_grad(self):
for name, param in self.model.named_parameters():
if param.requires_grad:
self.grad_backup[name] = param.grad.clone()
def restore_grad(self):
for name, param in self.model.named_parameters():
if param.requires_grad:
param.grad = self.grad_backup[name]
使用的时候，要麻烦一点：
pgd = PGD(model)
K = 3
for batch_input, batch_label in data:
# 正常训练
loss = model(batch_input, batch_label)
loss.backward() # 反向传播，得到正常的grad
pgd.backup_grad()
# 对抗训练
for t in range(K):
pgd.attack(is_first_attack=(t==0)) # 在embedding上添加对抗扰动, first attack时备份param.data
if t != K-1:
model.zero_grad()
else:
pgd.restore_grad()
loss_adv = model(batch_input, batch_label)
loss_adv.backward() # 反向传播，并在正常的grad基础上，累加对抗训练的梯度
pgd.restore() # 恢复embedding参数# 梯度下降，更新参数
optimizer.step()
model.zero_grad()
- hk2s: Traditional Chinese (Hong Kong standard) to Simplified Chinese
- s2hk: Simplified Chinese to Traditional Chinese (Hong Kong standard)
- s2t: Simplified Chinese to Traditional Chinese
- s2tw: Simplified Chinese to Traditional Chinese (Taiwan standard)
- s2twp: Simplified Chinese to Traditional Chinese (Taiwan standard, with phrases)
- t2hk: Traditional Chinese to Traditional Chinese (Hong Kong standard)
- t2s: Traditional Chinese to Simplified Chinese
- t2tw: Traditional Chinese to Traditional Chinese (Taiwan standard)
- tw2s: Traditional Chinese (Taiwan standard) to Simplified Chinese
- tw2sp: Traditional Chinese (Taiwan standard) to Simplified Chinese (with phrases)
3、DALL·E 2 can make realistic edits to existing images from a natural language caption. It can add and remove elements while taking shadows, reflections, and textures into account.
右边的图是左边的放大版，纵轴是GLUE分数，横轴是FLOPs (floating point operations)，Tensorflow中提供的浮点数计算量统计。从上图可以看到，同等量级的ELECTRA是一直碾压BERT的，而且在训练更长的步数之后，达到了当时的SOTA模型——RoBERTa的效果。从左图曲线上也可以看到，ELECTRA效果还有继续上升的空间
2. 模型结构
NLP式的Generator-Discriminator
ELECTRA最主要的贡献是提出了新的预训练任务和框架，把生成式的Masked language model(MLM)预训练任务改成了判别式的Replaced token detection(RTD)任务，判断当前token是否被语言模型替换过。那么问题来了，我随机替换一些输入中的字词，再让BERT去预测是否替换过可以吗？可以的，因为我就这么做过，但效果并不好，因为随机替换太简单了。
class BertTokenizer(PreTrainedTokenizer):
"""
Construct a BERT tokenizer. Based on WordPiece.
This tokenizer inherits from :class:`~transformers.PreTrainedTokenizer` which contains most of the main methods.
Users should refer to this superclass for more information regarding those methods.
...
"""
class BertModel(BertPreTrainedModel):
"""
The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
cross-attention is added between the self-attention layers, following the architecture described in `Attention is
all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
To behave as an decoder the model needs to be initialized with the :obj:`is_decoder` argument of the configuration
set to :obj:`True`. To be used in a Seq2Seq model, the model needs to initialized with both :obj:`is_decoder`
argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
input to the forward pass.
"""
# BertEncoder的前向传播返回部分，即上面的encoder_outputs
if not return_dict:
return tuple(
v
for v in [
hidden_states,
next_decoder_cache,
all_hidden_states,
all_self_attentions,
all_cross_attentions,
]
if v is not None
)
return BaseModelOutputWithPastAndCrossAttentions(
last_hidden_state=hidden_states,
past_key_values=next_decoder_cache,
hidden_states=all_hidden_states,
attentions=all_self_attentions,
cross_attentions=all_cross_attentions,
)
def apply_chunking_to_forward(
forward_fn: Callable[..., torch.Tensor], chunk_size: int, chunk_dim: int, *input_tensors
) -> torch.Tensor:
"""
This function chunks the :obj:`input_tensors` into smaller input tensor parts of size :obj:`chunk_size` over the
dimension :obj:`chunk_dim`. It then applies a layer :obj:`forward_fn` to each chunk independently to save memory.
If the :obj:`forward_fn` is independent across the :obj:`chunk_dim` this function will yield the same result as
directly applying :obj:`forward_fn` to :obj:`input_tensors`.
...
"""
补充：原论文中多头的理由为Multi-head attention allows the model to jointly attend to information from different representation subspaces at different positions. With a single attention head, averaging inhibits this.而另一个比较靠谱的分析有：
# ...
attention_scores = attention_scores / math.sqrt(self.attention_head_size)
if attention_mask is not None:
# Apply the attention mask is (precomputed for all layers in BertModel forward() function)
attention_scores = attention_scores + attention_mask # 这里为什么是+而不是*？# Normalize the attention scores to probabilities.
attention_probs = nn.Softmax(dim=-1)(attention_scores)
# This is actually dropping out entire tokens to attend to, which might# seem a bit unusual, but is taken from the original Transformer paper.
attention_probs = self.dropout(attention_probs)
# Mask heads if we want to
if head_mask is not None:
attention_probs = attention_probs * head_mask
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
# 省略decoder返回值部分……
return outputs
def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device) -> Tensor:
"""
Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
Arguments:
attention_mask (:obj:`torch.Tensor`):
Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
input_shape (:obj:`Tuple[int]`):
The shape of the input to the model.
device: (:obj:`torch.device`):
The device of the input to the model.
Returns:
:obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
"""
# 省略一部分……# Since attention_mask is 1.0 for positions we want to attend and 0.0 for# masked positions, this operation will create a tensor which is 0.0 for# positions we want to attend and -10000.0 for masked positions.# Since we are adding it to the raw scores before the softmax, this is# effectively the same as removing these entirely.
extended_attention_mask = extended_attention_mask.to(dtype=self.dtype) # fp16 compatibility
extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
return extended_attention_mask
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]# ourselves in which case we just need to make it broadcastable to all heads.
extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
class BertPooler(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.activation = nn.Tanh()
def forward(self, hidden_states):
# We "pool" the model by simply taking the hidden state corresponding# to the first token.
first_token_tensor = hidden_states[:, 0]
pooled_output = self.dense(first_token_tensor)
pooled_output = self.activation(pooled_output)
return pooled_output
class BertLMPredictionHead(nn.Module):
def __init__(self, config):
super().__init__()
self.transform = BertPredictionHeadTransform(config)
# The output weights are the same as the input embeddings, but there is# an output-only bias for each token.
self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
# Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
self.decoder.bias = self.bias
def forward(self, hidden_states):
hidden_states = self.transform(hidden_states)
hidden_states = self.decoder(hidden_states)
return hidden_states
不止BERT，所有huggingface实现的PLM的word embedding和masked language model的预测权重在初始化过程中都是共享的：
class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin):
# ...
def tie_weights(self):
"""
Tie the weights between the input embeddings and the output embeddings.
If the :obj:`torchscript` flag is set in the configuration, can't handle parameter sharing so we are cloning
the weights instead.
"""
output_embeddings = self.get_output_embeddings()
if output_embeddings is not None and self.config.tie_word_embeddings:
self._tie_or_clone_weights(output_embeddings, self.get_input_embeddings())
if self.config.is_encoder_decoder and self.config.tie_encoder_decoder:
if hasattr(self, self.base_model_prefix):
self = getattr(self, self.base_model_prefix)
self._tie_encoder_decoder_weights(self.encoder, self.decoder, self.base_model_prefix)
# ...
# model: a Bert-based-model object# learning_rate: default 2e-5 for text classification
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [
{'params': [p for n, p in param_optimizer if not any(
nd in n for nd in no_decay)], 'weight_decay': 0.01},
{'params': [p for n, p in param_optimizer if any(
nd in n for nd in no_decay)], 'weight_decay': 0.0}
]
optimizer = AdamW(optimizer_grouped_parameters,
lr=learning_rate)
# ...
2017年，Attention Is All You Need论文首次提出了Transformer模型结构并在机器翻译任务上取得了The State of the Art(SOTA, 最好)的效果。2018年，BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding使用Transformer模型结构进行大规模语言模型（language model）预训练（Pre-train），再在多个NLP下游（downstream）任务中进行微调（Finetune），一举刷新了各大NLP任务的榜单最高分，轰动一时。2019年-2021年，研究人员将Transformer这种模型结构和预训练+微调这种训练方式相结合，提出了一系列Transformer模型结构、训练方式的改进（比如transformer-xl，XLnet，Roberta等等）。如下图所示，各类Transformer的改进不断涌现。
