The "sys_probs" parameter doesn't seem to be working properly

[96XinHe]

deepmd-kit version: 3.0.0b4

I performed a single-task training process with the DPA-2 descriptor and specified "sys_probs" for each of the 28 training sets in "training":

"training": {
"stat_file": "./dpa2.hdf5",
"training_data": {
"systems": [
...............................
],
"batch_size": 1,
"sys_probs": [0.03,0.05,0.05,0.03,0.03,0.03,0.05,0.05,0.03,0.03,0.03,0.03,0.05,0.03,0.03,0.03,0.03,0.03,0.03,0.05,0.03,0.03,0.03,0.05,0.03,0.03,0.05,0.03],
"_comment": "that's all"
},

But it looks like this parameter doesn't work properly, and the probabilities for each training set in the output log are still assigned by default:

The “sys_probs” parameter is also correctly recognized in the auto-generated input_v2_compat.json file

[wanghan-iapcm]

which back-end did you use? or plz provide your command line

[96XinHe]

the command line is:
dp --pt train input_torch_medium.json

[iProzd]

sys_probs was not used in pt, because its priority was lower than that of auto_prob, while auto_prob always has its default values.
Fixed in #4353