我试图在 ppc64le 架构中使用torchmeta。不幸的是，安装并不容易，因为 ppc64le 需要特殊的二进制文件才能工作。
我最终设法按照以下说明为 pytorch 和 torchvision 获取了正确的二进制文件（在正确的 ibm 通道前面加上 conda 二进制文件，并安装了所有必需的文件）：
conda config --prepend channels https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda/
conda create -n my_new_env python=3.7 powerai=1.7.0
conda activate my_new_env
之后我开始安装正确版本的 torchmeta，1.3.1因为 ppc64le 只有 pytorch1.3.1和 torchvision 0.4.2。所以我做了：
pip install torchmeta==1.3.1
但现在我有一个新错误，它找不到与我想做的兼容的正确版本的 h5py。错误消息太大，无法粘贴，但我将粘贴我希望有用的部分：
(my_new_env) [miranda9@hal-login ~]$ pip install torchmeta==1.3.1
Collecting torchmeta==1.3.1
Using cached torchmeta-1.3.1-py3-none-any.whl (144 kB)
Requirement already satisfied: requests in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from torchmeta==1.3.1) (2.22.0)
Requirement already satisfied: torchvision<0.6.0,>=0.4.0 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from torchmeta==1.3.1) (0.4.2)
Requirement already satisfied: torch<1.5.0,>=1.3.0 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from torchmeta==1.3.1) (1.3.1)
Processing ./.cache/pip/wheels/87/f5/ad/9f04a48453875e8054c19f9fe3f50cbbe0c09b956835555019/Pillow-6.2.2-cp37-cp37m-linux_ppc64le.whl
Requirement already satisfied: numpy>=1.14.0 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from torchmeta==1.3.1) (1.17.4)
Requirement already satisfied: tqdm>=4.0.0 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from torchmeta==1.3.1) (4.36.1)
Collecting h5py~=2.9.0
Using cached h5py-2.9.0.tar.gz (287 kB)
Requirement already satisfied: chardet<3.1.0,>=3.0.2 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from requests->torchmeta==1.3.1) (3.0.4)
Requirement already satisfied: certifi>=2017.4.17 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from requests->torchmeta==1.3.1) (2020.6.20)
Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from requests->torchmeta==1.3.1) (1.25.10)
Requirement already satisfied: idna<2.9,>=2.5 in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from requests->torchmeta==1.3.1) (2.8)
Requirement already satisfied: six in ./.conda/envs/my_new_env/lib/python3.7/site-packages (from torchvision<0.6.0,>=0.4.0->torchmeta==1.3.1) (1.13.0)
Building wheels for collected packages: h5py
Building wheel for h5py (setup.py) ... error
ERROR: Command errored out with exit status 1:
command: /home/miranda9/.conda/envs/my_new_env/bin/python -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-install-bpmeop26/h5py/setup.py'"'"'; __file__='"'"'/tmp/pip-install-bpmeop26/h5py/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' bdist_wheel -d /tmp/pip-wheel-ccg1oj0n
cwd: /tmp/pip-install-bpmeop26/h5py/
Complete output (1321 lines):
running bdist_wheel
running build
running build_py
creating build
creating build/lib.linux-ppc64le-3.7
creating build/lib.linux-ppc64le-3.7/h5py
copying h5py/__init__.py -> build/lib.linux-ppc64le-3.7/h5py
copying h5py/h5py_warnings.py -> build/lib.linux-ppc64le-3.7/h5py
copying h5py/highlevel.py -> build/lib.linux-ppc64le-3.7/h5py
copying h5py/ipy_completer.py -> build/lib.linux-ppc64le-3.7/h5py
copying h5py/version.py -> build/lib.linux-ppc64le-3.7/h5py
creating build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/__init__.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/attrs.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/base.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/compat.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/dataset.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/datatype.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/dims.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/files.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/filters.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/group.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/selections.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/selections2.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
copying h5py/_hl/vds.py -> build/lib.linux-ppc64le-3.7/h5py/_hl
creating build/lib.linux-ppc64le-3.7/h5py/tests
copying h5py/tests/__init__.py -> build/lib.linux-ppc64le-3.7/h5py/tests
copying h5py/tests/common.py -> build/lib.linux-ppc64le-3.7/h5py/tests
creating build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/__init__.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_attrs.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_attrs_data.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_base.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_dataset.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_datatype.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_dimension_scales.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_file.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_file_image.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_group.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_h5.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_h5d_direct_chunk_write.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_h5f.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_h5p.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_h5t.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_objects.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_selections.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
copying h5py/tests/old/test_slicing.py -> build/lib.linux-ppc64le-3.7/h5py/tests/old
creating build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/__init__.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_attribute_create.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_dataset_getitem.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_dataset_swmr.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_datatype.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_deprecation.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_dims_dimensionproxy.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_file.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_filters.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
copying h5py/tests/hl/test_threads.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl
creating build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
copying h5py/tests/hl/test_vds/__init__.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
copying h5py/tests/hl/test_vds/test_highlevel_vds.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
copying h5py/tests/hl/test_vds/test_lowlevel_vds.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
copying h5py/tests/hl/test_vds/test_virtual_source.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
running build_ext
Autodetected HDF5 1.10.2
********************************************************************************
Summary of the h5py configuration
Path to HDF5: None
HDF5 Version: '1.10.2'
MPI Enabled: False
Rebuild Required: True
********************************************************************************
Executing api_gen rebuild of defs
Executing cythonize()
[ 1/22] Cythonizing /tmp/pip-install-bpmeop26/h5py/h5py/_conv.pyx
/tmp/pip-install-bpmeop26/h5py/.eggs/Cython-0.29.21-py3.7.egg/Cython/Compiler/Main.py:369: FutureWarning: Cython directive 'language_level' not set, using 2 for now (Py2). This will change in a later release! File: /tmp/pip-install-bpmeop26/h5py/h5py/_conv.pxd
tree = Parsing.p_module(s, pxd, full_module_name)
...
/home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h:17:2: warning: #warning "Using deprecated NumPy API, disable it with " "#define NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION" [-Wcpp]
#warning "Using deprecated NumPy API, disable it with " \
^
In file included from /tmp/pip-install-bpmeop26/h5py/h5py/defs.c:654:0:
/tmp/pip-install-bpmeop26/h5py/h5py/api_compat.h:27:18: fatal error: hdf5.h: No such file or directory
#include "hdf5.h"
^
compilation terminated.
error: command 'gcc' failed with exit status 1
----------------------------------------
ERROR: Failed building wheel for h5py
Running setup.py clean for h5py
Failed to build h5py
DEPRECATION: Could not build wheels for h5py which do not use PEP 517. pip will fall back to legacy 'setup.py install' for these. pip 21.0 will remove support for this functionality. A possible replacement is to fix the wheel build issue reported above. You can find discussion regarding this at https://github.com/pypa/pip/issues/8368.
Installing collected packages: Pillow, h5py, torchmeta
Attempting uninstall: Pillow
Found existing installation: Pillow 7.1.2
Uninstalling Pillow-7.1.2:
Successfully uninstalled Pillow-7.1.2
Attempting uninstall: h5py
Found existing installation: h5py 2.8.0
Uninstalling h5py-2.8.0:
Successfully uninstalled h5py-2.8.0
Running setup.py install for h5py ... error
ERROR: Command errored out with exit status 1:
command: /home/miranda9/.conda/envs/my_new_env/bin/python -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-install-bpmeop26/h5py/setup.py'"'"'; __file__='"'"'/tmp/pip-install-bpmeop26/h5py/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record /tmp/pip-record-hlwpfooj/install-record.txt --single-version-externally-managed --compile --install-headers /home/miranda9/.conda/envs/my_new_env/include/python3.7m/h5py
...
copying h5py/tests/hl/test_vds/test_lowlevel_vds.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
copying h5py/tests/hl/test_vds/test_virtual_source.py -> build/lib.linux-ppc64le-3.7/h5py/tests/hl/test_vds
running build_ext
Autodetected HDF5 1.10.2
********************************************************************************
Summary of the h5py configuration
Path to HDF5: None
HDF5 Version: '1.10.2'
MPI Enabled: False
Rebuild Required: True
********************************************************************************
Executing cythonize()
[ 1/22] Cythonizing /tmp/pip-install-bpmeop26/h5py/h5py/_conv.pyx
/tmp/pip-install-bpmeop26/h5py/.eggs/Cython-0.29.21-py3.7.egg/Cython/Compiler/Main.py:369: FutureWarning: Cython directive 'language_level' not set, using 2 for now (Py2). This will change in a later release! File: /tmp/pip-install-bpmeop26/h5py/h5py/_conv.pxd
...
warning: h5py/api_types_hdf5.pxd:730:6: 'H5Z_ERROR_EDC' redeclared
warning: h5py/api_types_hdf5.pxd:731:6: 'H5Z_DISABLE_EDC' redeclared
warning: h5py/api_types_hdf5.pxd:732:6: 'H5Z_ENABLE_EDC' redeclared
warning: h5py/api_types_hdf5.pxd:733:6: 'H5Z_NO_EDC' redeclared
building 'h5py.defs' extension
creating build/temp.linux-ppc64le-3.7
creating build/temp.linux-ppc64le-3.7/tmp
creating build/temp.linux-ppc64le-3.7/tmp/pip-install-bpmeop26
creating build/temp.linux-ppc64le-3.7/tmp/pip-install-bpmeop26/h5py
creating build/temp.linux-ppc64le-3.7/tmp/pip-install-bpmeop26/h5py/h5py
gcc -pthread -B /home/miranda9/.conda/envs/my_new_env/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DH5_USE_16_API -I./h5py -I/tmp/pip-install-bpmeop26/h5py/lzf -I/opt/local/include -I/usr/local/include -I/home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/numpy/core/include -I/home/miranda9/.conda/envs/my_new_env/include/python3.7m -c /tmp/pip-install-bpmeop26/h5py/h5py/defs.c -o build/temp.linux-ppc64le-3.7/tmp/pip-install-bpmeop26/h5py/h5py/defs.o
In file included from /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/numpy/core/include/numpy/ndarraytypes.h:1830:0,
from /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/numpy/core/include/numpy/ndarrayobject.h:12,
from /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/numpy/core/include/numpy/arrayobject.h:4,
from /tmp/pip-install-bpmeop26/h5py/h5py/api_compat.h:26,
from /tmp/pip-install-bpmeop26/h5py/h5py/defs.c:654:
/home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h:17:2: warning: #warning "Using deprecated NumPy API, disable it with " "#define NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION" [-Wcpp]
#warning "Using deprecated NumPy API, disable it with " \
^
In file included from /tmp/pip-install-bpmeop26/h5py/h5py/defs.c:654:0:
/tmp/pip-install-bpmeop26/h5py/h5py/api_compat.h:27:18: fatal error: hdf5.h: No such file or directory
#include "hdf5.h"
^
compilation terminated.
error: command 'gcc' failed with exit status 1
----------------------------------------
Rolling back uninstall of h5py
Moving to /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/h5py
from /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/~5py
Moving to /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/h5py-2.8.0-py3.7.egg-info
from /home/miranda9/.conda/envs/my_new_env/lib/python3.7/site-packages/~5py-2.8.0-py3.7.egg-info
ERROR: Command errored out with exit status 1: /home/miranda9/.conda/envs/my_new_env/bin/python -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-install-bpmeop26/h5py/setup.py'"'"'; __file__='"'"'/tmp/pip-install-bpmeop26/h5py/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record /tmp/pip-record-hlwpfooj/install-record.txt --single-version-externally-managed --compile --install-headers /home/miranda9/.conda/envs/my_new_env/include/python3.7m/h5py Check the logs for full command output.
任何人都知道如何在 ppc64le 中成功安装工作的 torchmeta 版本（使用 wmcle 1.7.0）？
有关的：
torchmeta 的 gitissue：https ://github.com/tristandeleu/pytorch-meta/issues/95
IBM gitissue 用于 torchmeta 支持：https ://github.com/IBM/powerai/issues/269
用于 torchmeta 的 h5py gitissue：https ://github.com/h5py/h5py/issues/1678
IBM h5py 对 torchmeta 的支持：https ://github.com/IBM/powerai/issues/270