v2.2.0-beta.0

New features

• support C API by @njzjz in #1098
  • add get cutoff, numb_types, and type_map to C API by @njzjz in #2061
  • c: support neighbor list input in the C API by @njzjz in #2106
  • c: support DeepPot model deviation in the C API by @njzjz in #2112
  • c: fix ComputeNlist -> ComputeNList in the C API by @njzjz in #2119
  • c: fix bug in hpp float interface and add tests by @njzjz in #2128
  • c: DeepTensor by @njzjz in #2184
  • add a header-only C++ DeepPot class based on C library by @njzjz in #2079
• support the multi-task training api by @iProzd in #2093
• support the finetune api by @iProzd in #1946
• supports left hands volume by @njzjz in #1956
• support bfloat16 by @njzjz in #1970
• support DPRc + type embedding by @njzjz in #2166

Enhancements

Python

• dp test output files: energy per atom, virial per atom by @likefallwind in #2029
• log HIP_VISIBLE_DEVICES for ROCm by @njzjz in #2032
• move run_config.ini into the root directory by @njzjz in #2098
• skip get_gpus subprocess when TF is cpu only by @njzjz in #2135
• add Optional to type hints when default is None by @njzjz in #2139
• optimize FioBin.save by @njzjz in #2143
• open the graph file only once in compress by @njzjz in #2144
• refactor enable_compression by @njzjz in #2147
• store energy bias with interface precision by @njzjz in #2174

Core

• dynamically load CUDA runtime library by @njzjz in #2082

C++

• uncouple model precision and interface precision by @njzjz in #2031
• merge double and float cc library by @njzjz in #2033
• add shuffle_exclude_empty to DeepPot.compute(nlist, atomic) by @njzjz in #2111

OP

• remove OP_LIB from op_abi by @njzjz in #2150

LAMMPS

• lmp: throw errors when 6-element atomic virial requested by @njzjz in #1983
• enable ttm without a patch by @njzjz in #2055
• bump lammps to stable_23Jun2022_update2 by @njzjz in #2053

Build and release

• CMake
  • use CPP_CXX_ABI for Python libraries by @njzjz in #1960
  • find protobuf headers in an automatic way by @njzjz in #1975
  • search TF deps from LD_LIBRARY_PATH; eliminate errors by @njzjz in #1976
  • CMake config
    • export c++ library to cmake config by @njzjz in #2049
    • export C library to cmake config by @njzjz in #2059
  • refactor USE_TF_PYTHON_LIBS by @njzjz in #2177
  • add an option to enable native compilation optimization by @njzjz in #2151
• Python and wheel building
  • migrate cibuildwheel configs to pyproject.toml and add tests by @njzjz in #2000
  • migrate from setup.cfg to pyproject.toml by @njzjz in #2011
  • migrate to PEP 517 and PEP 621 by @njzjz in #2035
  • store TF version in the wheel's extras_require by @njzjz in #2084
  • fix win64 builds and add win64 wheels by @njzjz in #2085
  • enable CUDA for linux-64 wheels by @njzjz in #2087
  • add macos-64 wheels by @njzjz in #2088
  • pin manylinux image to a specific tag for GCC 11 by @njzjz in #2137
  • retag wheels to py37-none-<platform> by @njzjz in #2180
  • build wheels for Linux ARM64 by @njzjz in #2179
  • pip install lammps module by @njzjz in #2186
• build_tf.py
  • build_tf.py: support ROCm; bump to TF 2.10 by @njzjz in #2030
  • build_tf.py: export LD_LIBRARY_PATH when executing configure by @njzjz in #2185
• build, package, and release C library by @njzjz in #2072
• perceive LAMMPS version for built-in mode by @njzjz in #1950

Test

• enable gcov for Python tests by @njzjz in #1969
• add unittests for LAMMPS pair deepmd by @njzjz in #1977
• c++: add clang tests and fix clang errors by @njzjz in #1992
• refactor: merge tests into the main cmake project by @njzjz in #2102
• merge double and float cc tests by @njzjz in #2108
• add ut for dp convert-from by @njzjz in #2175
• speed up the UTs
  • print out the time of each Python unit test by @njzjz in #2116
  • speed up test_neighbor_stat by @njzjz in #2120
  • speed up tests with CLI by @njzjz in #2121
• CI
  • upgrade ubuntu-18.04 by @njzjz in #2096
  • add pass step to GitHub Actions by @njzjz in #2141
  • give pass actions different names by @njzjz in #2142

Code cleanup

• clean deprecated codes by @njzjz in #2140

Documents

• docs: document lammps keyword relative_v, aparam, and ttm by @njzjz in #2060
• add dp test help doc by @likefallwind in #2038
• docs: fix HDF5 path by @njzjz in #2076
• docs: document pip install by @njzjz in #2091
• docs: document the conda-forge channel by @njzjz in #2095
• docs: fix spellings and grammars by @njzjz in #2114
• docs: rewrite docs for parallelism by @njzjz in #2099
• docs: add a FAQ about precision by @njzjz in #2115
• docs: add a note about lammps runpath by @njzjz in #2156
• the se-2 descriptor is multi-body by @amcadmus in #2069
• add header only library example by @njzjz in #2083
• Update python.md by @LiangWenshuo1118 in #2158

Bug fixings

• fix gcov coverage by @njzjz in #1951
• Fix typo of radisu by @kmu in #1954
• fix compatible with importlib_metadata 5.0.0 and Python 3.12 by @njzjz in #1961
• fix googletest download by @njzjz in #1949
• fix typo in the doc of loc_frame by @amcadmus in #1963
• replace : by ; from LD_LIBRARY_PATH in cmake by @njzjz in #1982
• CVE-2007-4559 Patch by @TrellixVulnTeam in #1984
• use float/double constants for spline5_switch by @njzjz in #1985
• remove duplicated convert_13_to_21 by @njzjz in #1986
• resolve macos clang errors by @njzjz in #1993
• fix Findtensorflow when LD_LIBRARY_PATH is empty by @njzjz in #2010
• fix OP_CXX_ABI for TF>=2.9 by @njzjz in #2013
• Fix compatibility with ROCm>=3.5.1; fix typo in hip neighbor_list by @njzjz in #2012
• fix python inference on windows by @njzjz in #2054
• cmake: set TensorFlow as import libraries by @njzjz in #2056
• build_tf.py: mirror fixes by @njzjz in #2057
• build_tf.py: roll back TF version to 2.9 by @njzjz in #2065
• fix type hints errors by @Angel-Jia in #2070
• improve building wheel workflow by @njzjz in #2074
• skip checking _GLIBCXX_USE_CXX11_ABI on windows or osx by @njzjz in #2075
• fix codecov by @njzjz in #2097
• fix TF compile warnings about Status::OK by @njzjz in #2109
• Fix version 0.12 model conversion by @Yi-FanLi in #2110
• fix invalid escape sequence in strings by @njzjz in #2122
• add precompile headers by @njzjz in #2123
• fix the missing header in api_c tests by @njzjz in #2131
• do not apply automatic batch size on CPUs by @njzjz in #2148
• pass user defined CUDA_NVCC_FLAGS and HIP_HIPCC_FLAGS by @njzjz in #2168

New Contributors

• @kmu made their first contribution in #1954
• @TrellixVulnTeam made their first contribution in #1984
• @Angel-Jia made their first contribution in #2070
• @LiangWenshuo1118 made their first contribution in #2158

Full Changelog: v2.1.5...v2.2.0.b0