Update DM21 pinned requirements and bazel config for compiling functional to C++.

PiperOrigin-RevId: 505176617

density_functional_approximation_dm21/.bazelversion

@@ -0,0 +1 @@
+5.3.0

density_functional_approximation_dm21/README.md

@@ -113,15 +113,14 @@ be found in the paper (reference below). Note that the results in our paper also
 include D3 corrections, which must be
 [included separately](https://pyscf.org/user/dft.html#dispersion-corrections).
 
-
 ### Best practices for using the neural functionals.
 
 In this section, we suggest some tips for using the neural functionals in a way
 similar to how they were used for benchmarking in the paper. The tensorflow
-network that we used is running at single precision, and as such it is very
-hard to converge calculations to the high convergence thresholds which are
-default in pyscf. For example, the following script should allow users to
-run an atomization energy calculation for methane.
+network that we used is running at single precision, and as such it is very hard
+to converge calculations to the high convergence thresholds which are default in
+pyscf. For example, the following script should allow users to run an
+atomization energy calculation for methane.
 
 ```python
 import density_functional_approximation_dm21 as dm21
@@ -178,17 +177,16 @@ print({'CH4': energies[0], 'C': energies[1], 'H': energies[2]})
 ```
 
 This script should produce three energies (in Hartrees) for the water molecule
-and the two atoms of
-`{'CH4': -40.51785372584538, 'C': -37.84542045526023, 'H': -0.5011533955627797}`
-, this leads to an atomization energy of 419.06 kcal/mol, which is then
-corrected with the D3(BJ) correction for methane (1.20 kcal/mol) to yield a
-predicted atomization energy of 420.26 kcal/mol. Comparing this to the
-literature value of 420.42, leads us to deduce an error of around 0.2 kcal/mol.
-
-It should also be noted that if a closed shell system is run unrestricted it
-can give a small difference between spin densities and eigenvalues with a
-negligible effect on the energy.
+and the two atoms of `{'CH4': -40.51785372584538, 'C': -37.84542045526023, 'H':
+-0.5011533955627797}` , this leads to an atomization energy of 419.06 kcal/mol,
+which is then corrected with the D3(BJ) correction for methane (1.20 kcal/mol)
+to yield a predicted atomization energy of 420.26 kcal/mol. Comparing this to
+the literature value of 420.42, leads us to deduce an error of around 0.2
+kcal/mol.
 
+It should also be noted that if a closed shell system is run unrestricted it can
+give a small difference between spin densities and eigenvalues with a negligible
+effect on the energy.
 
 ## Using DM21 from C++
 
@@ -210,10 +208,10 @@ library is provided in `cc/dm21_aot_compiled_example.cc`. This requires a
 link-time dependency on parts of the `xla_compiled_cpu_runtime_standalone`
 library, which are not included in the compiled functional library. The easiest
 way to build this is to use [Bazel](https://bazel.build). The first step is to
-[install Bazel](https://docs.bazel.build/versions/4.2.0/install.html).
+[install Bazel](https://docs.bazel.build/versions/5.3.0/install.html).
 [Bazelisk](https://docs.bazel.build/versions/main/install-bazelisk.html) is
 another way to install Bazel if a native installer is not available. The
-following has been tested with Bazel 4.2.0. It is best to continue working
+following has been tested with Bazel 5.3.0. It is best to continue working
 inside a virtual environment.
 
 Assuming the above installation steps using `git clone` have been followed, and
@@ -221,7 +219,7 @@ Assuming the above installation steps using `git clone` have been followed, and
 
 ```
 pip install -r requirements_aot_compilation.txt
-bazel run --experimental_repo_remote_exec :run_dm21_aot_compiled_example
+bazel run --experimental_cc_shared_library --experimental_repo_remote_exec :run_dm21_aot_compiled_example
 ```
 
 where the `pip install` command is only required if a fresh virtual environment

density_functional_approximation_dm21/WORKSPACE.bazel

@@ -20,25 +20,16 @@ pip_install(
 
 http_archive(
     name = "io_abseil_py",
-    strip_prefix = "abseil-py-pypi-v0.15.0",
-    urls = ["https://github.com/abseil/abseil-py/archive/pypi-v0.15.0.tar.gz"],
-)
-
-http_archive(
-    name = "six_archive",
-    build_file = "@io_abseil_py//third_party:six.BUILD",
-    strip_prefix = "six-1.12.0",
-    urls = [
-        "https://pypi.python.org/packages/source/s/six/six-1.12.0.tar.gz",
-    ],
+    strip_prefix = "abseil-py-1.4.0",
+    urls = ["https://github.com/abseil/abseil-py/archive/refs/tags/v1.4.0.tar.gz"],
 )
 
 http_archive(
     name = "org_tensorflow",
     patch_args = ["-p1"],
     patches = ["tf_bazel.patch"],
-    strip_prefix = "tensorflow-2.10.0",
-    urls = ["https://github.com/tensorflow/tensorflow/archive/refs/tags/v2.10.0.tar.gz"],
+    strip_prefix = "tensorflow-2.11.0",
+    urls = ["https://github.com/tensorflow/tensorflow/archive/refs/tags/v2.11.0.tar.gz"],
 )
 
 # The cascade of load() statements and tf_workspace?() calls works around the

density_functional_approximation_dm21/external/tf_bazel.patch

@@ -6,23 +6,36 @@ Subject: [PATCH] Set dependencies for saved_model_compile_aot rule,
  tensorflow, and absl version.
 
 ---
- tensorflow/compiler/tf2xla/BUILD  | 2 +-
- tensorflow/python/tools/tools.bzl | 4 ++--
- 2 files changed, 3 insertions(+), 3 deletions(-)
+ tensorflow/compiler/tf2xla/BUILD              | 2 +-
+ tensorflow/compiler/xla/mlir/ir/runtime/BUILD | 1 +
+ tensorflow/python/tools/tools.bzl             | 4 ++--
+ 3 files changed, 4 insertions(+), 3 deletions(-)
 
 diff --git a/tensorflow/compiler/tf2xla/BUILD b/tensorflow/compiler/tf2xla/BUILD
 index cfe63b16675..cd273aaf29e 100644
 --- a/tensorflow/compiler/tf2xla/BUILD
 +++ b/tensorflow/compiler/tf2xla/BUILD
 @@ -235,7 +235,7 @@ cc_library(
-     copts = runtime_copts() + tf_openmp_copts(),
-     features = ["fully_static_link"],
+         "-parse_headers",
+     ],
      linkstatic = 1,
 -    visibility = [":friends"],
 +    visibility = ["//visibility:public"],
      # Note, we specifically removed MKL and multithreaded dependencies so the
      # standalone does not require the MKL binary blob or threading libraries.
      #
+diff --git a/tensorflow/compiler/xla/mlir/ir/runtime/BUILD b/tensorflow/compiler/xla/mlir/ir/runtime/BUILD
+index 8c693fa5d49..bf95a3de622 100644
+--- a/tensorflow/compiler/xla/mlir/ir/runtime/BUILD
++++ b/tensorflow/compiler/xla/mlir/ir/runtime/BUILD
+@@ -90,6 +90,7 @@ cc_library(
+         "rt_ops.h",
+     ],
+     compatible_with = get_compatible_with_cloud(),
++    visibility = ["//visibility:public"],
+     deps = [
+         ":rt_inc_gen",
+         "//tensorflow/compiler/xla/runtime:constraints",
 diff --git a/tensorflow/python/tools/tools.bzl b/tensorflow/python/tools/tools.bzl
 index db886746006..bc597e29de9 100644
 --- a/tensorflow/python/tools/tools.bzl

density_functional_approximation_dm21/requirements.txt

@@ -1,9 +1,9 @@
-absl-py==0.13.0
-attrs==21.2.0
-h5py==3.7.0
-numpy==1.23.4
+absl-py==1.4.0
+attrs==22.2.0
+h5py==3.8.0
+numpy==1.24.1
 pyscf==2.1.1
 pytest==6.2.4
 scipy==1.9.3
-tensorflow==2.10.0
+tensorflow==2.11.0
 tensorflow-hub==0.12.0

density_functional_approximation_dm21/requirements_aot_compilation.txt

@@ -1,3 +1,3 @@
 Keras-Preprocessing==1.1.2
-numpy==1.23.4
-tensorflow-estimator==2.10.0
+numpy==1.24.1
+tensorflow-estimator==2.11.0