Why Is the Performance of the Fine-Tuned Pangu Model Unsatisfactory When the Data Volume Is Sufficient?

Locate the fault as follows:

• Check the quality of training data. The model training effect will be compromised if training data or its distribution does not align with the target task, training data contains abnormal data, or data diversity is poor. If these problems occur, improve the data quality.