Finetuning on a chatbot responses dataset

[P-Sood]

Hello, I just wanted to start off and say this is wonderful work!

I am creating a chatbot, and this chatbot dataset is in this form:

messages=[
      {
        "role": "system",
        "content": prompt,
      },
      {
          "role": "assistant",
          "content": "AI: Message"
      },
      {
          "role": "user",
          "content": "CX: Message"
      },
      {
          "role": "assistant",
          "content": "AI: Message"
      },
      {
          "role": "user",
          "content": "CX: Message"
      },
      {
          "role": "assistant",
          "content": "AI: Message"
      },
      {
          "role": "user",
          "content": "CX: Message"
      },
      {
          "role": "assistant",
          "content": "AI: Message"
      },
      {
          "role": "user",
          "content": "CX: Message"
      },
      {
          "role": "assistant",
          "content": "AI: Message"
      },
      {
          "role": "user",
          "content": "CX: Message"
      },
    ],

Where it shows a direct conversation clearly between the user and the AI assistant. I have looked at your service/modelservice.py page and was wondering if the function OpenAIChatModelProvider is what I would need to change in order to accept my dataset. Currently, it just takes all input as just the users.

Moreover, just wondering if there are some general issues you may see arise from using this dataset.

Cheers,