search

tensorflow / fold

Deep learning with dynamic computation graphs in TensorFlow
Apache License 2.0
1.83k stars 266 forks source link

can current fold work with tensorflow 1.3? Is it tested? #85

Open xiaoyunwu opened 7 years ago

evelkey commented 7 years ago

No, I tried it and found out some cuda ops are not compatible. For quick solution I merged different versions of tensorflow to have the functionality I need.

delesley commented 7 years ago

Can you be more specific about your problems? The binaries are outdated, but when compiling from source, it compiles and runs fine with 1.3 on my machine. You may have an issue with checksums when downloading tensorflow dependencies. If you run git log, there's a link to a bug report that tells you how to hack a .bzl file to fix it.

On Thu, Sep 28, 2017 at 5:52 AM, Géza Velkey notifications@github.com wrote:

No, I tried it and found out some cuda ops are not compatible. For quick solution I merged different versions of tensorflow to have the functionality I need.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tensorflow/fold/issues/85#issuecomment-332826713, or mute the thread https://github.com/notifications/unsubscribe-auth/AGGbTaqQActNzkzzfdGa17gJwtU1yliJks5sm5aWgaJpZM4PnI9I .

-- DeLesley Hutchins | Software Engineer | delesley@google.com | 505-206-0315

KazuhiraDZ commented 7 years ago

Hi, I want to compile from source, can you tell me how to compile the latest version? is it the same as how we compile tensorflow from source?

evelkey commented 7 years ago

Hi, after compiling both tensorflow_fold and tensorflow from sources, it is working fine with tf version 1.3. It's quite tricky to get it to work, because the tensorflow version which is included in fold has an error with a checksum. You can find build instructions here: sources Right now it's building with only bazel 5.4 for me. To make it build you have to remove the checksum from tf with the following command: sed -ri "/^\W+sha256 = \"[^\"]+\"\W+$/d" tensorflow/workspace.bzl (issue)

KazuhiraDZ commented 7 years ago

thank you for help. I have run the code to remove the checksum,but, it still has something wrong,like: Python Configuration Error: 'PYTHON_BIN_PATH' environment variable is not set and referenced by '@org_tensorflow//util/python:python_headers' ERROR: Analysis of target '//tensorflow_fold/loom:pywrapweaver' failed; build aborted INFO: Elapsed time: 1.325s FAILED: Build did NOT complete successfully (8 packages loaded) Fetching http://mirror.bazel.build/.../swig/swig-3.0.8/swig-3.0.8.tar.gz Fetching http://mirror.bazel.build/.../eigen/eigen/get/f3a22f35b044.tar.gz Fetching http://mirror.bazel.build/.../backports.weakref-1.0rc1.tar.gz Fetching http://mirror.bazel.build/.../archive/release-1.8.0.zip ERROR: Couldn't start the build. Unable to run tests can you help me the order how to build it? and every time it have different log

evelkey commented 7 years ago

Try exporting the variable replacing "usr/bin/python3" with your own python path: export PYTHON_BIN_PATH="/usr/bin/python3"

KazuhiraDZ commented 7 years ago

Thank you! It does work in cpu pc! I will install it in my gpu server...

shengc commented 6 years ago

I have the following errors trying to build from source. Any idea how to fix them ?

/usr/include/string.h:544: error: '__locale_t' has not been declared
/usr/include/string.h:545: error: nonnull argument references non-pointer operand (argument 1, operand 3)
/usr/include/string.h:548: error: '__locale_t' has not been declared
/usr/include/string.h:549: error: nonnull argument references non-pointer operand (argument 1, operand 4)