Deep potential plugin for CP2K implemented from C interface by Cloudac7 · Pull Request #3145 · cp2k/cp2k

Cloudac7 · 2023-12-06T14:47:37Z

Enhanced based on #1620
For the implementation is fully changed, I think it would be better to just open a new PR instead of former old one.

Deep Potential (DP) is a machine learning potential based on deep neural network, first proposed by Zhang L. et al. (Doi: 10.1016/j.cpc.2018.03.016) By learning from coordinates as well as energy and forces of different structures from DFT calculations, a potential describing the whole system could be frozen to a file. By reading the file, program could yield energy, forces etc. from structures, and then MD simulation with DFT accuracy could be fastened greatly. This PR add Deep Potential support for CP2K, with potential file (e.g. graph.pb) supplied, CP2K could run MD simulation.

As mentioned in #1620 , it was not a good practice to introduce C++ flags in CP2K, while at that time DeePMD-kit does not supply C interface. But now everything has changed. Since v2.2.x, DeePMD-kit C Interface is directly released together with each DeePMD-kit release, making it no need to install libtensorflow_cc any more. In this way, C wrapper has been rewritten and now only necessary parameters are added to arch file.

We also support download from toolchain. To enable the DeePMD-kit feature, --with-deepmd flag in toolchain should not be set to no. The following ways installing has been added.

Please set --deepmd-mode to gpu if GPU support is needed, and load CUDA environment before installing.
From exist DeePMD-kit C interface, by setting --with-deepmd=/path/to/libdeepmd_c.
From source code and library from Github, by setting it to install. For now, DeePMD-kit v2.2.7 will be installed.
From system, by adding the header files and library to where ld command could read directly.

src/deepmd_wrapper.F

fstein93 · 2023-12-06T15:48:09Z

src/fist_nonbond_env_types.F

+            DEALLOCATE (fist_nonbond_env%deepmd_data%force)
+         END IF
+         IF (ASSOCIATED(fist_nonbond_env%deepmd_data%use_indices)) THEN
+            DEALLOCATE (fist_nonbond_env%deepmd_data%use_indices)


I would refactor that into a routine in the DeepMD module

fstein93 · 2023-12-06T15:49:27Z

src/manybody_deepmd.F

+      ALLOCATE (use_atom(n_atoms))
+      ALLOCATE (use_atom_type(n_atoms))
+      use_atom = .FALSE.
+      use_atom_type = 100 ! just initial the array


src/manybody_deepmd.F

src/pair_potential_types.F

fstein93 · 2023-12-06T15:54:28Z

src/pair_potential_types.F

+   SUBROUTINE pair_potential_deepmd_copy(deepmd_source, deepmd_dest)
+      TYPE(deepmd_pot_type), POINTER                     :: deepmd_source, deepmd_dest
+
+      IF (.NOT. ASSOCIATED(deepmd_source)) RETURN


I would rather nullify deepmd_dest

INSTALL.md

oschuett · 2023-12-07T12:10:16Z

It's so nice to see DeePMD land after all :-)

This PR looks very good to me. My only concern is with the H2O-DPLR-model.pb file because its 10.3 MB. Is there maybe a smaller model we could use for testing?

Cloudac7 · 2023-12-08T06:58:03Z

It's so nice to see DeePMD land after all :-)

This PR looks very good to me. My only concern is with the H2O-DPLR-model.pb file because its 10.3 MB. Is there maybe a smaller model we could use for testing?

I just replaced the model as one of 1.6 MB in AISSquare. It could be a rather small one to reproduce results just similar to calculation result for a W cell as shown in materials project.

tools/toolchain/scripts/stage6/install_deepmd.sh

src/common/bibliography.F

tools/toolchain/install_cp2k_toolchain.sh

Make code Pretty adapt to libdeepmd_c fix dummy argument of BIND(C) fix some bug in compilation

test: add regtest for DeePMD

fix: fix compiling when no `-D__DEEPMD` style: reformat INSTALL document

test: use W-DP-HYB-model as test case

oschuett · 2024-01-04T12:36:51Z

Seems like there is a memory leak. Can you fix this in DeepMD? Otherwise you could add a suppression.

Cloudac7 · 2024-01-04T14:29:45Z

Seems like there is a memory leak. Can you fix this in DeepMD? Otherwise you could add a suppression.

As discussed in deepmodeling/deepmd-kit#3045, there is memory leak in Tensorflow and Protobuf, which could not be so easily fixed, while @njzjz has fixed a possible leak in DeePMD-kit (deepmodeling/deepmd-kit#3062, thanks a lot.)

At this moment, I think just adding suppression might be OK while the leak is not so severe that might cause the program crashing.

robinzyb · 2024-01-04T14:36:01Z

It is nice to see this interface land!

oschuett · 2024-01-04T15:35:23Z

I'm also very happy to finally merge this. Great work!

Btw, we now have a section on machine learning in our manual ;-)

njzjz · 2024-01-04T15:45:54Z

Thanks, @Cloudac7.

DeePMD-kit also has a documentation page for all external interfaces.

fstein93 reviewed Dec 6, 2023

View reviewed changes

INSTALL.md Outdated Show resolved Hide resolved

njzjz reviewed Dec 8, 2023

View reviewed changes

tools/toolchain/scripts/stage6/install_deepmd.sh Outdated Show resolved Hide resolved

njzjz reviewed Dec 8, 2023

View reviewed changes

tools/toolchain/scripts/stage6/install_deepmd.sh Outdated Show resolved Hide resolved

njzjz reviewed Dec 8, 2023

View reviewed changes

src/common/bibliography.F Outdated Show resolved Hide resolved

njzjz reviewed Dec 14, 2023

View reviewed changes

tools/toolchain/install_cp2k_toolchain.sh Outdated Show resolved Hide resolved

Cloudac7 added 17 commits January 4, 2024 15:01

Add DeePMD-kit C++ Interface

e7a2ad0

Make code Pretty adapt to libdeepmd_c fix dummy argument of BIND(C) fix some bug in compilation

fix: invalid read and write of memory

3b31027

style: make code pretty

32a66d8

chore: disable shellcheck

46a8d4f

fix: wrong parameter names

632d905

test: add regtest for DeePMD

refactor: redesign input section

ba76aef

fix: fix compiling when no `-D__DEEPMD` style: reformat INSTALL document

fix: invalid virial in sdbg

831de26

test: use W-DP-HYB-model as test case

test: fix result of regtest

817d3e0

fix: double counting from multiple MPI processes

5a0678f

chore: reformat reference

4c24d27

fix: 'c_null_ptr' has no IMPLICIT type

a456a70

style: make code pretty

037f568

chore: update biography;

34e7abc

chore: remove ldeepmd_op

46f3f6a

fix: index error for more than 2 elements

8fbd71a

chore: remove useless tfcc and deepmd_mode

26a0164

chore: make code pretty

bc6def1

Cloudac7 force-pushed the deepmd-c-interface branch from 8d1452d to bc6def1 Compare January 4, 2024 07:04

style: make code pretty

08f830f

test: add lsan.supp for deepmd and tf

ebc5dc9

oschuett merged commit 969d126 into cp2k:master Jan 4, 2024

Cloudac7 mentioned this pull request Jan 12, 2024

doc: add DeePMD-kit document #3213

Merged

oschuett mentioned this pull request Jan 19, 2024

Toolchain: Add leak suppression for DeePMD #3233

Merged

Conversation

Cloudac7 commented Dec 6, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

fstein93 Dec 6, 2023

Choose a reason for hiding this comment

Uh oh!

fstein93 Dec 6, 2023

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

fstein93 Dec 6, 2023

Choose a reason for hiding this comment

Uh oh!

Uh oh!

oschuett commented Dec 7, 2023

Uh oh!

Cloudac7 commented Dec 8, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

oschuett commented Jan 4, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Cloudac7 commented Jan 4, 2024

Uh oh!

robinzyb commented Jan 4, 2024

Uh oh!

oschuett commented Jan 4, 2024

Uh oh!

njzjz commented Jan 4, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Cloudac7 commented Dec 6, 2023 •

edited

Loading

Cloudac7 commented Dec 8, 2023 •

edited

Loading

oschuett commented Jan 4, 2024 •

edited

Loading