TF 1.15 is very old which does to consider the changes of Numpy ABI and updates after python 3.10.
So we need to fix those problems to make sure TF 1.15 is well supported for existing customers.
The difficulty of changes of Numpy since 1.19 requires us to have tensorflow build numpy 1.19 by applying changes in numpy side.
However python 3.10 does not work with numpy < 1.24, due to numpy ABI changes which requires more changes from numpy side (unlike to apply patches upon python + Cython)
Soluiton
Two relase for Tf1.15:
Tf1.15+py3.8/py3.9 : this release is used for tensorflow 1.15 with python <= 3.9, numpy < 1.19.0
Tf1.15+py3.10/py3.12: this release is used for latest optimizatoin could be applied upon tf1.15 with python >= 3.10, numpy >= 1.24
Both versions have fixed protobuf 3.8.0 but we need make sure protobuf works well with python3.x.
Previously we only have ROC6.1 image for python 3.10 and tensorflow 2.x, for the successive release we should have
Description
TF 1.15 is very old which does to consider the changes of Numpy ABI and updates after python 3.10.
So we need to fix those problems to make sure TF 1.15 is well supported for existing customers.
The difficulty of changes of Numpy since 1.19 requires us to have tensorflow build numpy 1.19 by applying changes in numpy side.
However python 3.10 does not work with numpy < 1.24, due to numpy ABI changes which requires more changes from numpy side (unlike to apply patches upon python + Cython)
Soluiton
Two relase for Tf1.15:
Tf1.15+py3.8/py3.9 : this release is used for tensorflow 1.15 with python <= 3.9, numpy < 1.19.0
Tf1.15+py3.10/py3.12: this release is used for latest optimizatoin could be applied upon tf1.15 with python >= 3.10, numpy >= 1.24
Both versions have fixed protobuf 3.8.0 but we need make sure protobuf works well with python3.x.
Previously we only have ROC6.1 image for python 3.10 and tensorflow 2.x, for the successive release we should have