ERROR on proc 0: Substitution for illegal variable NSTEPS (../input.cpp:651)

[SHANG-Ming-Hui]

Dear all,
Can you kindly give me some guidence for addressing the error encountered during the model_devi step ?
The error message is ** ERROR on proc 0: Substitution for illegal variable NSTEPS (../input.cpp:651)
Last command: run ${NSTEPS} upto **.
The number of processes in a task is the number of times this error appeares.

My environment: deepmd-kit-2.2.4-gpu version coupled with LAMMPS-stable_2Aug2023
DP-GEN-0.10.4 and dpdata-0.2.4 were both installed.

Model can be trained using the "dpgen run" flow. While the "model_devi" could not be performed due to the error raised as bellow:
ERROR on proc 0: Substitution for illegal variable NSTEPS (../input.cpp:651)
Last command: run ${NSTEPS} upto

The corresponding log file was attached.
The parameter.json for model_devi step is set as:

---

"model_devi_dt": 0.001,
"model_devi_skip": 0,
"model_devi_f_trust_lo": 0.05,
"model_devi_f_trust_hi": 0.15,
"model_devi_e_trust_lo": 10000000000.0,
"model_devi_e_trust_hi": 10000000000.0,
"model_devi_clean_traj": true,
"model_devi_jobs": [
  {"sys_idx": [0], "temps": [500, 380, 300], "press": [1.0], "trj_freq": 10, "nsteps": 2000, "ensemble": "nvt", "_idx": "00"}

---

The machine.json for model_devi step is set as:

---

"machine": {
"context_type": "local",
"batch_type": "slurm",
"local_root": "./",
"remote_root": "xxxxxxxxxxxxxxxxxxxxxx(path)/gputrain-2/tmp/md",
"_comment": "that's all"
},
"resources": {
"number_node": 1,
"cpu_per_node": 48,
"gpu_per_node": 4,
"queue_name": "gpu",
"strategy":{"if_cuda_multi_devices":true},
"custom_flags": [
"#SBATCH -J LMP",
"#SBATCH -N 1",
"#SBATCH --ntasks-per-node=4",
"#SBATCH --ntasks-per-socket=2",
"#SBATCH --cpus-per-task=10",
"#SBATCH -t 24:00:00",
"#SBATCH --mem=32G"
],
"source_list": [
"/public/home/smh/software/lmp.env"
],
"module_list": [],
"_time_limit": "",
"group_size": 5
},
"command": "mpirun -np 4 ~/software/lammps-stable_2Aug2023/bin/lmp_mpi -sf gpu -pk gpu 1 tpa 8"

---

Best wishes
12/1, 2023
model_devi.log

[njzjz]

The error is caused by the LAMMPS script and should be not related to DeePMD-kit.

[njzjz]

This is a version mismatching. DP_DeepPotIsAParamNAll is a new symbol, and an old DeePMD-kit library might be loaded.

export LD_DEBUG=libs will help to find which library is actually loaded instead of the correct one.

[SHANG-Ming-Hui]

Hi, Jinzhe, thanks so much.
Currently, i found the model_deviation can be performed by manually changing the "run ${NSTEPS} upto" as "run 2000 upto" in the last line of input.lammps generated within the "dpgen run" work flow. Thus the lmp_mpi seems to be compiled successfuly.
Does it imply that the interface between DeepMD-kit and lammps was incorrectly build ?
Thanks again.

[njzjz]

I don't understand why ${NSTEPS} didn't work.

It's also unclear why it worked with the CPU version.

[njzjz]

You may post the entire LAMMPS input file.

[SHANG-Ming-Hui]

The inputs for LAMMPS running with error:

---

`plugin load libdeepmd_lmp.sovariable NSTEPS equal 1000 ##### Normally, a line break is required. However, there is no line break in the current file, which leads to an error in variable setting.
variable THERMO_FREQ equal 10
variable DUMP_FREQ equal 10
variable TEMP equal 500.000000
variable PRES equal -1.000000
variable TAU_T equal 0.100000
variable TAU_P equal 0.500000

units metal
boundary p p p
atom_style atomic

neighbor 1.0 bin

box tilt large
if "${restart} > 0" then "read_restart dpgen.restart.*" else "read_data conf.lmp"
change_box all triclinic
mass 1 133.000000
mass 2 119.000000
mass 3 127.000000
pair_style deepmd ../graph.000.pb ../graph.001.pb ../graph.002.pb ../graph.003.pb out_freq ${THERMO_FREQ} out_file model_devi.out
pair_coeff * *

thermo_style custom step temp pe ke etotal press vol lx ly lz xy xz yz
thermo ${THERMO_FREQ}
dump 1 all custom ${DUMP_FREQ} traj/*.lammpstrj id type x y z fx fy fz
restart 10000 dpgen.restart

if "${restart} == 0" then "velocity all create ${TEMP} 786184"
fix 1 all nvt temp ${TEMP} ${TEMP} ${TAU_T}

timestep 0.001000
run ${NSTEPS} upto`

---

The inputs for LAMMPS works well:
`plugin load libdeepmd_lmp.sovariable NSTEPS equal 1000 ##### Normally, a line break is required. However, there is no line break in the current file, which leads to an error in variable setting.
variable THERMO_FREQ equal 10
variable DUMP_FREQ equal 10
variable TEMP equal 500.000000
variable PRES equal -1.000000
variable TAU_T equal 0.100000
variable TAU_P equal 0.500000

units metal
boundary p p p
atom_style atomic

neighbor 1.0 bin

box tilt large
if "${restart} > 0" then "read_restart dpgen.restart.*" else "read_data conf.lmp"
change_box all triclinic
mass 1 133.000000
mass 2 119.000000
mass 3 127.000000
pair_style deepmd ../graph.000.pb ../graph.001.pb ../graph.002.pb ../graph.003.pb out_freq ${THERMO_FREQ} out_file model_devi.out
pair_coeff * *

thermo_style custom step temp pe ke etotal press vol lx ly lz xy xz yz
thermo ${THERMO_FREQ}
dump 1 all custom ${DUMP_FREQ} traj/*.lammpstrj id type x y z fx fy fz
restart 10000 dpgen.restart

if "${restart} == 0" then "velocity all create ${TEMP} 786184"
fix 1 all nvt temp ${TEMP} ${TEMP} ${TAU_T}

timestep 0.001000
run 2000 upto`

---

For the CPU-version, LAMMPS is not loaded as the GPU-version. So the LAMMPS task can be perfomed normally by DeepMD-kit-CPU, while failed in the GPU version.
Perhaps, the problem can be resolved by modifying the code or something else. I'm now trying to do this.

[SHANG-Ming-Hui]

[SHANG-Ming-Hui]

Presently, the problem can been solved by adding a line break at the end of the statement "ret = "plugin load libdeepmd_lmp.so"" within the dpgen/generator/lib/lammps.py-plugin.
Which should be changed as:

---

ret = "plugin load libdeepmd_lmp.so \n" #### adding a line break.

---