【经验总结】超算互联网服务器 transformers 加载本地模型

1. 背景

使用 超算互联网 的云服务，不能连接外网，只能把模型下载到本地，再上传上去到云服务。

2. 模型下载

在 模型中 https://huggingface.co/models 找到所需的模型后

点击下载 config.json pytorch_model.bin vocab.txt

3. 上传模型文件

打开文件管理 e-file
点击 上传，选择 本地上传

4. 运行代码

由原来的代码：

from transformers import BertTokenizer, BertForSequenceClassification

model = BertForSequenceClassification.from_pretrained('bert-base-chinese', num_labels=2)
tokenizer = BertTokenizer.from_pretrained('bert-base-chinese')

改成这样的：

from transformers import BertTokenizer, BertForSequenceClassification


path_to_local_model_directory="/public/home/acc5trotmy/jupyter/models/bert-base-chinese"

model = BertForSequenceClassification.from_pretrained(path_to_local_model_directory, num_labels=2)
tokenizer = BertTokenizer.from_pretrained(path_to_local_model_directory)

或者：

from transformers import BertTokenizer, BertForSequenceClassification, BertConfig

config = BertConfig.from_pretrained(path_to_local_model_directory)
model = BertForSequenceClassification.from_pretrained(path_to_local_model_directory, config=config)
tokenizer = BertTokenizer.from_pretrained(path_to_local_model_directory)