deepmd-kit 2.0.2, error of init-frz-model command

[zjd11]

I use init-frz-model command to train a frozen model generated by this command, and i find the newly trained model is the same as the old one, because i use the two models to perform a MD calculation, the result is same. I have tested different "auto_prob_style": "prob_sys_size; when i train the new model. How can i fix the error.

[njzjz]

Did you freeze the model again?

[zjd11]

yes, when i finished the train, i freeze the model again.

[njzjz]

Sorry, I cannot reproduce your issue. Could you provide an example for reproducing?

[zjd11]

I have uploaded an example, including data set, input.json, and so on. I find, in deepmd-kit-2.0.2, i use 'dp train -ini-freeze-model' command to train a model based on a compressed one, the obtained model is the same as the original one，because the lammps output is the same. but when i use the command to train a model based on a uncompressed one, the results are different.
202.zip

[njzjz]

I can reproduce the phenomenon. cc @denghuilu

[damonturney]

Was this error fixed?

[zjd11]

yes, when i finished the train, i freeze the model again.