about V2.0.0 and V1.3.3. Successly compress model but program goes wrong while running lammps

[zhao-w-en]

Dear all:
Recently I have install V2.0.0 and try to run lammps with compressed model. At first I used uncompressed model like
pair_style deepmd frozen_model.pb
pair_coeff * *
, but when I submit the lammps task I met error like this
Invalid argument: Tensor model_attr/model_version:0, specified in either feed_devices or fetch_devices was not found in the Graph

Invalid argument: Tensor model_attr/model_version:0, specified in either feed_devices or fetch_devices was not found in the GraphInvalid argument: Tensor model_attr/model_version:0, specified in either feed_devices or fetch_devices was not found in the Graph

APPLICATION TERMINATED WITH THE EXIT STRING: Aborted (signal 6)

terminate called after throwing an instance of 'std::runtime_error' what(): incompatable model: version 0.0 in graph, but version 1.0 supported

terminate called after throwing an instance of 'terminate called after throwing an instance of 'std::runtime_error'

Then I changed the in.lammps as
pair_style deepmd frozen_model_compress.pb
pair_coeff * *
and the error changed like this
Internal: 2 root error(s) found.
(0) Internal: Blas xGEMMStridedBatched launch failed : a.shape=[14,4,200], b.shape=[14,4,4], m=200, n=4, k=4, batch_size=14
[[{{node filter_type_1/MatMul}}]]
[[o_force/_31]]
(1) Internal: Blas xGEMMStridedBatched launch failed : a.shape=[14,4,200], b.shape=[14,4,4], m=200, n=4, k=4, batch_size=14
[[{{node filter_type_1/MatMul}}]]
0 successful operations.
0 derived errors ignored.
I have no idea about these error, by the way, lammps will work properly use v1.3.3 if I use uncompressed dp model. I'm really confusing. Could you please give me some advice?
Waitting for your reply

[WU2JING]

Hello, you can try to run lammps with V2.0.0. The compressed model suit for V2.0.0.

[zhao-w-en]

Thank you so much!

[ZCY10410]

Hi, do you fix this problem successfully? I meet the same problem when I run lammps with V2.0.0. I am testing a deep potential trained in published work. I guess the potential I test is trained with DeepMD-kit version 1.

Waiting for your reply.

[JTaozhang]

I meet the same problem, "INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the GraphINVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph". My deepmd-kit version is v2.2.2, lmp is 23 Jun 2022.

[njzjz]

This model is likely generated by an old DeePMD-kit version. Are you sure the version is v2.2.2?

[JTaozhang]

yes, I am sure. I listed the initial part of log file here. LAMMPS (23 Jun 2022 - Update 1)
using 1 OpenMP thread(s) per MPI task
Loaded 1 plugins from /share/home/zhangtao/anaconda3/envs/deepmd2/lib/deepmd_lmp
Reading data file ...
triclinic box = (0 0 0) to (23.52804 20.37588 25) with tilt (-11.76402 1.5e-15 2.7e-15)
WARNING: Triclinic box skew is large (src/domain.cpp:224)
8 by 14 by 1 MPI processor grid
reading atoms ...
364 atoms
Finding 1-2 1-3 1-4 neighbors ...
special bond factors lj: 0 0 0
special bond factors coul: 0 0 0
0 = max # of 1-2 neighbors
0 = max # of 1-3 neighbors
0 = max # of 1-4 neighbors
1 = max # of special neighbors
special bonds CPU = 0.015 seconds
read_data CPU = 0.041 seconds
Summary of lammps deepmd module ...

Info of deepmd-kit:
installed to: /share/home/zhangtao/anaconda3/envs/deepmd2
source: v2.2.2
source branch: HEAD
source commit: 92ca097
source commit at: 2023-05-24 13:45:03 +0800
surpport model ver.:1.1
build variant: cpu
build with tf inc: /share/home/zhangtao/anaconda3/envs/deepmd2/include;/share/home/zhangtao/anaconda3/envs/deepmd2/include
build with tf lib: /share/home/zhangtao/anaconda3/envs/deepmd2/lib/libtensorflow_cc.so
set tf intra_op_parallelism_threads: 0
set tf inter_op_parallelism_threads: 0
Info of lammps module:
use deepmd-kit at: /share/home/zhangtao/anaconda3/envs/deepmd2INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph

[njzjz]

I see very little information as it does not show the version of the Python code. Did this message stop the program?

[JTaozhang]

Hi,
the output file doesn't contain any python information. I just run lammps. I don't think it has any connection with python. The message didn't stop the program. If you assign the program with 56 cores cpu, the messsage will repeat 56 times.

the start of the message like this,
LAMMPS (23 Jun 2022 - Update 1)
using 1 OpenMP thread(s) per MPI task
Loaded 1 plugins from /share/home/zhangtao/anaconda3/envs/deepmd2/lib/deepmd_lmp
Reading data file ...
triclinic box = (0 0 0) to (23.52804 20.37588 25) with tilt (-11.76402 1.5e-15 2.7e-15)
WARNING: Triclinic box skew is large (src/domain.cpp:224)
8 by 14 by 1 MPI processor grid
reading atoms ...
364 atoms
Finding 1-2 1-3 1-4 neighbors ...
special bond factors lj: 0 0 0
special bond factors coul: 0 0 0
0 = max # of 1-2 neighbors
0 = max # of 1-3 neighbors
0 = max # of 1-4 neighbors
1 = max # of special neighbors
special bonds CPU = 0.016 seconds
read_data CPU = 0.045 seconds
Summary of lammps deepmd module ...

Info of deepmd-kit:
installed to: /share/home/zhangtao/anaconda3/envs/deepmd2
source: v2.2.2
source branch: HEAD
source commit: 92ca097
source commit at: 2023-05-24 13:45:03 +0800
surpport model ver.:1.1
build variant: cpu
build with tf inc: /share/home/zhangtao/anaconda3/envs/deepmd2/include;/share/home/zhangtao/anaconda3/envs/deepmd2/include
build with tf lib: /share/home/zhangtao/anaconda3/envs/deepmd2/lib/libtensorflow_cc.so
set tf intra_op_parallelism_threads: 0
set tf inter_op_parallelism_threads: 0
Info of lammps module:
use deepmd-kit at: /share/home/zhangtao/anaconda3/envs/deepmd2INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph

the end of the message like this,
NVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph
INVALID_ARGUMENT: Tensor spin_attr/ntypes_spin:0, specified in either feed_devices or fetch_devices was not found in the Graph

Info of model(s):
using 1 model(s): graph.001.pb
rcut in model: 7
ntypes in model: 1

CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE

Your simulation uses code contributions which should be cited:

• USER-DEEPMD package:
  The log file lists these citations in BibTeX format.

CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE

WARNING: Using 'neigh_modify every 1 delay 0 check yes' setting during minimization (src/min.cpp:187)
Generated 0 of 0 mixed pair_coeff terms from geometric mixing rule
Neighbor list info ...
update every 1 steps, delay 0 steps, check yes
max neighbors/atom: 4000, page size: 200000
master list distance cutoff = 7.3
ghost atom cutoff = 7.3
binsize = 3.65, bins = 10 6 7
1 neighbor lists, perpetual/occasional/extra = 1 0 0
(1) pair deepmd, perpetual

Although the program report the message, the program still can be finished normally. But it is annoying to see these message.

[njzjz]

The Python interface generates the model, and it is essential to check if the Python and C++ interface versions match.

[JTaozhang]

Hi,

So, how to check the match between Python and C++ interface versions. and how can I solve this? I install deepmd-kit through conda command directly, I think the interface of Python and C++ woubld be matched automatically. can you give me some advice to solve this ?

best regards,

[njzjz]

The version is printed on the screen during training.

[JTaozhang]

Yes, I have seen the python version is 3.10. I also listed the lib version here.

Do those lib versions confict with the python ?