Dart如何解析OpenAI API返回的非标准JSON格式？

调用OpenAI API获取ChatGPT响应时，返回的是ByteStream。用utf8.decode将字节转成字符串后，会得到带data: 前缀的JSON片段，无法直接传入jsonDecode()解析，示例内容如下：

data: {"id":"chatcmpl-6rcqJGoHnJLz94ZNxcA569JsJSJsR","object":"chat.completion.chunk","created":1678238715,"model":"gpt-3.5-turbo-0301","choices":[{"delta":{"role":"assistant"},"index":0,"finish_reason":null}]}

完整的解码后内容是多行类似格式的字符串：

data: {"id":"chatcmpl-6rcqJGoHnJLz94ZNxcA569JsJSJsR","object":"chat.completion.chunk","created":1678238715,"model":"gpt-3.5-turbo-0301","choices":[{"delta":{"role":"assistant"},"index":0,"finish_reason":null}]}
data: {"id":"chatcmpl-6rcqJGoHnJLz94ZNxcA569JsJSJsR","object":"chat.completion.chunk","created":1678238715,"model":"gpt-3.5-turbo-0301","choices":[{"delta":{"role":"assistant"},"index":0,"finish_reason":null}]}

解决步骤 #

• 将解码后的字符串按行分割，过滤空行和结束标记data: [DONE]
• 对每一行去除开头的data: 前缀（注意可能存在的空格）
• 对处理后的纯JSON字符串调用jsonDecode()解析

示例代码（Dart） #

import 'dart:convert';

void processOpenAIStream(String decodedStream) {
  List<String> lines = decodedStream.split('\n');
  
  for (String line in lines) {
    String trimmed = line.trim();
    // 跳过空行和结束标记
    if (trimmed.isEmpty || trimmed == 'data: [DONE]') continue;
    // 移除data: 前缀
    String jsonPart = trimmed.replaceFirst(RegExp(r'^data: '), '');
    // 解析JSON
    Map<String, dynamic> chunk = jsonDecode(jsonPart);
    // 按需处理chunk数据，比如提取回复内容
    print(chunk['choices'][0]['delta']);
  }
}

内容的提问来源于stack exchange，提问作者Soliev