transformers bin保存模型为onnx、pt

参考：
https://huggingface.co/docs/transformers/v4.20.1/en/serialization#exporting-a-model-to-onnx

***默认模型保存地址在：~/.cache/huggingface
自定义读取保存地址cache_dir：

##指定读取
AutoModel.from_pretrained('bert-base-chinese', cache_dir='./')

#指定保存
# text_encoder.config.to_json_file(r'D:op***esave_model')
# text_tokenizer.save_vocabulary(r'D:ope***esave_model')

text_tokenizer.save_pretrained('local-pt-checkpoint')
text_encoder.save_pretrained('local-pt-checkpoint')

***只能保存模型，tokenizer单独加载

from transformers import BertForSequenceClassification, BertConfig, BertTokenizer

text_tokenizer = BertTokenizer.from_pretrained("IDEA-CCNL/Taiyi-CLIP-Roberta-102M-Chinese")
text_encoder = BertForSequenceClassification.from_pretrained("IDEA-CCNL/Taiyi-CLIP-Roberta-102M-Chinese").eval()

##save laod
import torch

torch.save(text_encoder, 'text_encoder.pt')

text_encoder1 = torch.load('text_encoder.pt')

## save onnx
text1 = text_tokenizer(["一只猫"])
torch.onnx.export(text_encoder1, torch.tensor(text1["input_ids"]), "text_encoder1.onnx",opset_version=10,
        )
        
#或添加return_tensors返回就是tensor
text1 = text_tokenizer(["一只猫"], return_tensors='pt', padding=True)["input_ids"]
torch.onnx.export(text_encoder1, text1, "text_encoder1.onnx",opset_version=10
        )

**其中torch.onnx.export报错RuntimeError: Unsupported: ONNX export of Slice with dynamic inputs. DynamicSlice is a deprecated
参考：https://blog.csdn.net/qq_29695701/article/details/115914898

***更改opset_version更高点，opset_version=10即可

前置需求：
1、pytorch版本需要1.8.0版本及以上
2、安装pip install transformers[onnx]

转化：

 python -m transformers.onnx --model="ID模型地址或者Model ID on huggingface.co"  D:op保存地址e_modelonnx