Which TF version to install?
vinay-hebb opened this issue · 3 comments
It is not clear which version of tensorflow to install for the code on top of branch? Is the code expected to run for any TF2.x? I could not see any requirements file to install compatible versions. Can any please let me know how to go ahead?
Minor versions should match. So install TF 2.10.x with TF Text 2.10.x.
If you are building a custom binary out of a branch like 2.9, any version of TF 2.9 or higher should work.
Minor versions should match. So install TF 2.10.x with TF Text 2.10.x.
If you are building a custom binary out of a branch like 2.9, any version of TF 2.9 or higher should work.
Sure, understood. Is there any specific reason why not to go with requirements.txt file (as it is general process followed) because it creates confusion about which version to use.
If you install using the pip package, this is maintained in install_requires.
If you are trying to build manually, you should be running the configure.sh and prepare_tf_dep.sh scripts (or run_build.sh which calls them). This is written in the install instructions in the main readme. These scripts provide more flexibility than requirements.txt because you can build against any flavor of TF you may have - tensorflow, tensorflow-cpu, tf-nightly, etc. There are also specific Bazel settings that are required to match your installed TF package that the scripts take care of as well.