transformers 获取中文字向量

代码如下：

# coding=utf-8
import torch
from transformers import BertTokenizer, BertModel

tokenizer = BertTokenizer.from_pretrained("D:/data/transformers/chinese_L-12_H-768_A-12")
model = BertModel.from_pretrained("D:/data/transformers/chinese_L-12_H-768_A-12")

token_ids = tokenizer.encode("我爱中国")
print(token_ids)
# [101, 2769, 4263, 704, 1744, 102]

tokens = tokenizer.decode(token_ids=token_ids)
print(tokens)
# [CLS] 我 爱 中 国 [SEP]

input_ids = torch.tensor(token_ids).unsqueeze(0)
print(input_ids)
# tensor([[ 101, 2769, 4263,  704, 1744,  102]])

outputs = model(input_ids)

# last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
sequence_output = outputs[0]
pooled_output = outputs[1]
print(sequence_output.shape)  ## 字向量
# torch.Size([1, 6, 768])
print(pooled_output.shape)  ## 句向量
# torch.Size([1, 768])

vector1 = sequence_output[:, 1].squeeze(0)  # '我' 字向量
vector2 = sequence_output[:, 2].squeeze(0)  # '爱' 字向量