Closed zfarrell13 closed 10 months ago
linking #878
@zfarrell13 Hello, tensorflow-text
relies on the build of tensorflow
, and starting from tensorflow 2.14
, it no longer supports Python 3.8
.
Hey @sun1638650145 , unfortunately I'm working on a project that requires 3.8. Ive found a precompiled macOS whl and got it working in 3.8. But, I cannot build a dockerfile for it as docker uses Linux under the hood. Hope that makes sense. Ive seen you post precompiled whl files elsewhere, if you know where I can find a 3.8 Linux whl, I'd be forever grateful.
@sun1638650145 to be clear, in this project, ive been able to execute tf gin models in python 3.8 by:
pip install tf pip install tensorflow_text-2.8.2-cp38-cp38-macosx_11_0_arm64.whl pip installing jax, jaxlib, etc...
what i cant do is build a dockerfile for my project.
when i try to pip install tensorflow_text-2.8.2-cp38-cp38-macosx_11_0_arm64.whl in the dockerfile, i get the following error:
ERROR: failed to solve: process "/bin/sh -c pip install tensorflow_text-2.8.2-cp38-cp38-macosx_11_0_arm64.whl" did not complete successfully: exit code: 1
which makes sense. i need a file for: tensorflow_text-2.8.2-cp38-cp38-manylinux_11_0_arm64.whl
in #878, there was a file for this but the link is inactive. Im confident i can get through the dependencies if i had this file, as ive worked throguh the same dependencies on macos.
@zfarrell13 I think I may understand your requirement now. If you need the manylinux2014_aarch64
version of tensorflow-text 2.8.2
, you should try building it yourself using the tensorflow-aarch64
version. Currently, only the x86-64
version of whl
for text
is officially released.
@sun1638650145 I was able to build a whl myself, but i could only figure out how to build an x86-64 version for python 3.8. probably just a matter of number of different combinations of build processes until i figure it out. in the meantime i built this colab for anyone else struggling: https://colab.research.google.com/drive/1sCAoRpH61cJhrwlDi-VZxpdqoisUUsKc?usp=sharing
Seems as though im still able to build a dockerfile with that using: docker build --platform linux/amd64 -t
see title. looking for a kind resource before i try and fail to compile myself.