Is FinGPTv1/2/3 made by finetuning General LLM (Llama or ChatGLM) with Language Modeling task, or Sentiment Analysis task, or both?

[ciaochiaociao]

After reading both the short and the long version of FinGPT paper, I am a little bit confused of how FinGPTv1/v2/v3 were really made. Were they made by taking a pre-trained model, e.g., Llama, ChatGLM, and fine-tuning on

1. Language Modeling task
2. Sentiment Analysis task

and a follow-up question is the dataset used for training was obtained by FinNLP real-time data API, right?

It would be great if you could provide more details into it.

I appreciate that.