train.json format

[wang-kangkang]

我查到openai.File.create的输入需要是jsonl格式，并且每一行是一个字典，它的key为’prompt’ ‘completion’，但我查看我生成的train.json，它的结构是: I found The input for openai.File.create() needs to be in JSONL format, and each line should be a dictionary with key 'prompt' and 'completion'. However, when I look at the trainjson I generated, its structure is: [ --{'messages', ----['role':'system', 'content':'.....'], ----['role':'user', 'content':'.....'] , ----['role':'assistant', 'content':'.....'], --}, --{'messages', ----['role':'system', 'content':'.....'], ----['role':'user', 'content':'.....'] , ----['role':'assistant', 'content':'.....'], --}, --{'messages', ----['role':'system', 'content':'.....'], ----['role':'user', 'content':'.....'] , ----['role':'assistant', 'content':'.....'], --},
...... ] 看起来这个结构的数据不能用啊，该怎么理解这个问题？ It seems that the data in this structure cannot be used. How should we understand this problem?