About model training and prompt construction

[Dojay6103]

As a beginner, I found that the effect of relationship extraction was not very satisfactory after training. I would like to ask you what prompts you use during testing or what techniques you have for training dialogue construction? In addition, will your data set of more than 10,000 items be open sourced in the future?

[nanoponge]

As a beginner, I found that the effect of relationship extraction was not very satisfactory after training. I would like to ask you what prompts you use during testing or what techniques you have for training dialogue construction? In addition, will your data set of more than 10,000 items be open sourced in the future?

The instructions during fine-tuning also incorporate similar prompt words as in the data generation process. In addition, the effectiveness of the model would be improved if the data on the relationship extraction in the dataset were then manually corrected. Retraining on some open source information extraction LLMs might also optimize the generation results.