要义 MLX 针对mac的ARM处理器优化的大模型训练架构

看了那么多资料主要还是PIP安装 huggingface_hub mlx-lm transformers torch numpy

然后对应大模型文件微调

针对数据 Completion（问答类型，一问一答） chat（角色问答） text（单文本类型，训练特定文本） 三种类型 都是JSON格式

chat解析：

{
  "messages": [
    {
      "role": "user",
      "content": "How do I use this product?"
    },
    {
      "role": "assistantA",
      "content": "To use this product, first insert the batteries and then press the power button."
    },
    {
      "role": "assistantB",
      "content": "To operate this product, make sure it's charged, and then follow the instructions in the manual."
    }
  ]
}

message包裹下为信息，role为user的content为询问，assistantB和assistantA给出不同的content

completion解析

[{
  "prompt": "What is the capital of France?",
  "completion": "Paris."
},
{
  "prompt": "A?",
  "completion": "BBBB"
}]

TEXT格式

{"text": "table: 1-1000181-1\ncolumns: State/territory, Text/background colour, Format, Current slogan, Current series, Notes\nQ: Tell me what the notes are for South Australia \nA: SELECT Notes FROM 1-1000181-1 WHERE Current slogan = 'SOUTH AUSTRALIA'"}