The fp process cannot iterate after dpgen training is complete

[Baopidragon1]

param1.json
I have three qustions.

1. My training set contains computational results from 300-8000K for multiple interface structures with the goal of making a reactive force field. However, at the end of the first training round of 200000 steps, rmse_f_trn only reaches the level of x.xe-01. Is this one caused by the training set?
2. After the training is completed and the model_devi process is run, the level of max_devi_f only reaches x.xe+00, at which point the number of failed structures is always 100 percent by setting the values of model_devi_f_trust_hi and model_devi_f_trust_lo, which are commonly used online. So I cranked up the model_devi_f_trust_hi and model_devi_f_trust_lo values to get the fp process to run. I was sure that the fp results were converging, but the results show no convergence. Here is a screenshot of the relevant input parameter file and error reporting. I hope some expert can help to solve the problem.

[param1.json](https://github.com/user-attachments/files/16043298/param1.json)

[njzjz]

It will be helpful to post a file that you believe converged but the program reported not.