Predictions of energies are all same!

[B-C-WANG]

After training for about 1000 epochs, the energy prediction of all coordinate is a Gaussian distribution, and after training for about 100000 epochs, the energy predictions are all same.
Here is the plot of prediction energy-real energy:

I think the input coordinate did not encode in a correct way, my system is a 3x3x3 Cu layer with 4 Co and 20 H2O, here is my input json file:

{
"_comment": " model parameters",
"use_smooth": true,
"sel_a": [50, 50,50,60,50],
"rcut_smth": 5.50,
"rcut": 5.80,
"filter_neuron": [25, 50, 100],
"filter_resnet_dt": false,
"n_axis_neuron": 16,
"n_neuron": [240, 240, 240],
"resnet_dt": true,
"_comment": " traing controls",
"systems": ["/opt/deepmd/test/Train"],
"set_prefix": "set",
"stop_batch": 100000,
"batch_size": 100,
"start_lr": 0.005,
"decay_steps": 5000,
"decay_rate": 0.95,
"start_pref_e": 0.02,
"limit_pref_e": 1,
"start_pref_f": 1000,
"limit_pref_f": 1,
"start_pref_v": 0,
"limit_pref_v": 0,
"seed": 1,
"_comment": " display and restart",
"_comment": " frequencies counted in batch",
"disp_file": "lcurve.out",
"disp_freq": 100,
"numb_test": 50,
"save_freq": 100,
"save_ckpt": "model.ckpt",
"load_ckpt": "model.ckpt",
"disp_training": true,
"time_training": true,
"profiling": false,
"profiling_file": "timeline.json",
"_comment": "that's all"
}

[amcadmus]

Please try
"batch_size": 1,

[B-C-WANG]

@amcadmus It works! Thanks!