The effect is worse after fine-tuning

[liyujia011025]

The effect is worse after fine-tuning

Hello expert:
I used my own data to fine tune the Moirai large model according to the fine-tuning process and code in the official README file, and found that the effect after fine-tuning was actually worse than without fine-tuning, which is strange.
Among them, because I set the context length to 96 and the prediction length to 4 during prediction, I set the parameters in the cli/conf/finetune/val_data/data.yaml file as shown in the figure. The learning rate in moirai-1.0-R-small.yaml was set to 1e-7, and other places such as hyperparameters were not changed. I tried fine-tuning all layers of Moirai and some layers of the output layer separately, but the results were even worse.
Do you know where my mistake occurred? Do I need to make specific changes to the hyperparameter settings or other content during the fine-tuning process?
Looking forward to your reply, thank you very much!