问题描述：

API访问返回信息

1、用MindIE服务化部署glm4-9b-chat模型，通过api访问，返回信息如下：

mindie-service日志报错如下：

>>> Exception:ChatGLM4Tokenizer._pad() got an unexpected keyword argument 'padding_side'

详细信息如下：

2025-03-19 20:38:30.895+08:00] [8742] [127.0.0.1:54920] [server] [request mindie server method:POST, uri:/v1/chat/completions] [success]
[2025-03-19 20:38:30.895+08:00] [8742] [9987] [mindie-server] [INFO] [http_server.cpp:49] : [endpoint] Receive request from 127.0.0.1:54920 ,method:POST, uri:/v1/chat/completions
[2025-03-19 20:38:30.896+08:00] [8742] [9987] [mindie-server] [DEBUG] [http_handler.cpp:1189] : [endpoint] OpenAI support by vLLM process.
[Tokenizer]        >>> Exception:ChatGLM4Tokenizer._pad() got an unexpected keyword argument 'padding_side'
[2025-03-19 20:38:30.921+08:00] [9949] [8743] [mindie-server] [ERROR] [infer_tokenizer.cpp:69] : [MIE04E04060B] [tokenizer] Encode chat fail: N8pybind1117error_already_setE, RuntimeError: [Tokenizer] encode chat template failed.

At:
  /usr/local/lib/python3.11/site-packages/mies_tokenizer/tokenizer.py(263): encode_chat

[2025-03-19 20:38:30.921+08:00] [8742] [9987] [mindie-server] [DEBUG] [infer_tokenizer.cpp:534] : [tokenizer] Encode prompt returns 0 tokens.

问题原因

tokenization_chatglm.py文件版本和mindie中不匹配

glm-4-9b-chat · 模型库

问题解决：

修改tokenization_chatglm.py脚本增加padding_side参数，或者下载最新的tokenization_chatglm.py脚本

参考：

https://blog.csdn.net/m0_60801087/article/details/143160274