Prefix-Tuning源码解析

Prefix-Tuning在PEFT包中的源码实现
改写自Based on https://github.com/THUDM/P-tuning-v2/blob/main/model/prefix_encoder.py

import torch
from transformers import PretrainedConfigclass PrefixEncoder(torch.nn.Module):r'''The torch.nn model to encode the prefixInput shape: (batch-size, prefix-length)Output shape: (batch-size, prefix-length, 2*layers*hidden)'''def __init__(self, config):super().__init__()self.prefix_projection = config.prefix_projectionif self.prefix_projection:# Use a two-layer MLP to encode the prefixself.embedding = torch.nn.Embedding(config.prefix_length, config.hidden_size)self.trans = torch.nn.Sequential(torch.nn.Linear(config.hidden_size, config.encoder_hidden_size),torch.nn.Tanh(),torch.nn.Linear(config.encoder_hidden_size, config.num_hidden_layers * 2 * config.hidden_size))else:self.embedding = torch.nn.Embedding(config.prefix_length, config.num_hidden_layers * 2 * config.hidden_size)def forward(self, prefix: torch.Tensor):if self.prefix_projection:prefix_tokens = self.embedding(prefix)past_key_values = self.trans(prefix_tokens)else:past_key_values = self.embedding(prefix)return past_key_valuesif __name__ == "__main__":configs = {"prefix_length":20,"hidden_size":768,"encoder_hidden_size":768,"num_hidden_layers":12,"prefix_projection":False}prefix_encoder = PrefixEncoder(config=PretrainedConfig.from_dict(configs))print(prefix_encoder)batch_size = 8prefix = torch.arange(20).long().expand(batch_size, -1)print(prefix.shape)output = prefix_encoder(prefix)print(output.shape)

下面我们以T5-large模型为例子：
不考虑Use a two-layer MLP to encode the prefix的话，prefix tuning主要包括以下代码：

class PrefixEncoder(torch.nn.Module):def __init__(self, config):super().__init__()...self.embedding = torch.nn.Embedding(num_virtual_tokens, num_layers * 2 * token_dim) #num_virtual_tokens=20,token_dim=1024,num_layers=24def forward(self, prefix: torch.Tensor):past_key_values = self.embedding(prefix)return past_key_values

得到的PrefixEncoder被传入peft->peft_model.py->prompt_encoder：

PrefixEncoder((embedding): Embedding(20, 49152) # 1024*2*24
)

self.prompt_tokens初始化为长度2*20的向量，因为T5有编码器和解码器，需要两次prefix：

self.prompt_tokens[adapter_name] = torch.arange(config.num_virtual_tokens * config.num_transformer_submodules).long() #20*2# tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 17,
#        18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
#        36, 37, 38, 39])

prompt_tokens = (self.prompt_tokens[self.active_adapter].unsqueeze(0).expand(batch_size, -1).to(prompt_encoder.embedding.weight.device)) 
prompt_tokens = prompt_tokens[:, : peft_config.num_virtual_tokens]
# 此时prompt_tokens.shape = (batch_size=8, num_virtual_tokens=20)past_key_values = prompt_encoder(prompt_tokens)
torch.Size([8, 20, 49152])

但目前的past_key_values还是所有层的集合，我们需要把past_key_values分解为每一层：

past_key_values = past_key_values.view(batch_size, #8peft_config.num_virtual_tokens, #20peft_config.num_layers * 2, #24*2peft_config.num_attention_heads, #16peft_config.token_dim // peft_config.num_attention_heads, #1024/16)
# torch.Size([8, 20, 48, 16, 64])

因为有编码器和解码器，所以再复制一次

past_key_values = torch.cat([past_key_values, past_key_values], dim=2)
# torch.Size([8, 20, 96, 16, 64])# 重排：torch.Size([96, 8, 16, 20, 64])
# 然后split成一个长度为24的tuple，每个tuple的shape：torch.Size([4, 8, 16, 20, 64])
past_key_values = past_key_values.permute([2, 0, 3, 1, 4]).split(peft_config.num_transformer_submodules * 2)

也就是说past_key_values是24个层的Prefix embedding，形状为`(num_transformer_submodules * 2, batch_size, num_attention_heads, num_virtual_tokens, token_dim/num_attention_heads])

注意这里*2是因为key+value.

transformers->models->t5->modeling_t5.py->T5Attention类，这里的关键步骤是project函数中的hidden_states = torch.cat([past_key_value, hidden_states], dim=2)，注意project函数仅仅用于key和value。

def forward(self,hidden_states,mask=None,key_value_states=None,position_bias=None,past_key_value=None,layer_head_mask=None,query_length=None,use_cache=False,output_attentions=False,):"""Self-attention (if key_value_states is None) or attention over source sentence (provided by key_value_states)."""# Input is (batch_size, seq_length, dim)# Mask is (batch_size, key_length) (non-causal) or (batch_size, key_length, key_length)# past_key_value[0] is (batch_size, n_heads, q_len - 1, dim_per_head)batch_size, seq_length = hidden_states.shape[:2]real_seq_length = seq_lengthif past_key_value is not None:if len(past_key_value) != 2:raise ValueError(f"past_key_value should have 2 past states: keys and values. Got { len(past_key_value)} past states")real_seq_length += past_key_value[0].shape[2] if query_length is None else query_lengthkey_length = real_seq_length if key_value_states is None else key_value_states.shape[1]def shape(states):"""projection"""return states.view(batch_size, -1, self.n_heads, self.key_value_proj_dim).transpose(1, 2)def unshape(states):"""reshape"""return states.transpose(1, 2).contiguous().view(batch_size, -1, self.inner_dim)def project(hidden_states, proj_layer, key_value_states, past_key_value):"""projects hidden states correctly to key/query states"""if key_value_states is None:# self-attn# (batch_size, n_heads, seq_length, dim_per_head)hidden_states = shape(proj_layer(hidden_states))elif past_key_value is None:# cross-attn# (batch_size, n_heads, seq_length, dim_per_head)hidden_states = shape(proj_layer(key_value_states))if past_key_value is not None:if key_value_states is None:# self-attn# (batch_size, n_heads, key_length, dim_per_head)# 注意这里是重点：用串联方式hidden_states = torch.cat([past_key_value, hidden_states], dim=2)elif past_key_value.shape[2] != key_value_states.shape[1]:# checking that the `	sequence_length` of the `past_key_value` is the same as# the provided `key_value_states` to support prefix tuning# cross-attn# (batch_size, n_heads, seq_length, dim_per_head)hidden_states = shape(proj_layer(key_value_states))else:# cross-attnhidden_states = past_key_valuereturn hidden_statesreal_seq_length += past_key_value[0].shape[2] if query_length is None else query_length

分别计算query_states、key_states、value_states，用query和key计算attention score，得到score形状为torch.Size([8, 16, 2, 22])，所以输入X可以attend to itself以及prefix。

    # hidden_states shape: torch.Size([8, 2, 1024])   # get query statesquery_states = shape(self.q(hidden_states))  # (batch_size, n_heads, seq_length, dim_per_head) # query_states shape: torch.Size([8, 16, 2, 64])# get key/value stateskey_states = project(hidden_states, self.k, key_value_states, past_key_value[0] if past_key_value is not None else None)# key_states shape: torch.Size([8, 16, 22, 64])value_states = project(hidden_states, self.v, key_value_states, past_key_value[1] if past_key_value is not None else None)# value_states shape: torch.Size([8, 16, 22, 64])# compute scores# torch.Size([8, 16, 2, 22])scores = torch.matmul(query_states, key_states.transpose(3, 2))  # equivalent of torch.einsum("bnqd,bnkd->bnqk", query_states, key_states), compatible with onnx op>9

接下来就是经典的attention操作了。用attn_weights ([8, 16, 2, 22]) 和value_states ([8, 16, 22, 64])相乘，把22消掉，就是每个输入X的输出了。

# if key and values are already calculated
# we want only the last query position bias
# position_bias.shape: torch.Size([8, 16, 2, 22])scores += position_bias_maskedattn_weights = nn.functional.softmax(scores.float(), dim=-1).type_as(scores)  # (batch_size, n_heads, seq_length, key_length)attn_weights = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)  # (batch_size, n_heads, seq_length, key_length)attn_output = unshape(torch.matmul(attn_weights, value_states))  # (batch_size, seq_length, dim) torch.Size([8, 2, 1024])attn_output = self.o(attn_output)present_key_value_state = (key_states, value_states) if (self.is_decoder and use_cache) else Noneoutputs = (attn_output,) + (present_key_value_state,) + (position_bias,)if output_attentions:outputs = outputs + (attn_weights,)return outputs

参考

https://huggingface.co/docs/peft/task_guides/seq2seq-prefix-tuning