Help me to understand final compilable python and tensorflow version

[vfr-xcugas]

Hi, Hope you are good there. Please help me to understand the version.

In google colab, we cant install in python 3.6, tensorflow 1.15.0

In conda, we can install 1.15.0 when python version 3.6

But, when I use tensorflow == 1.15.0 and run command bash install.sh
It throw error:
tensorflow/core/framework/op.h: No such file or directory
compilation terminated.

So, I tried install with Python 3.6 and tensorflow 2.4.0 using conda
It compite first step.

But for second step, when I use tensorflow 2.4.0 and python 3.6
I get following error:
AttributeError: module 'tensorflow' has no attribute 'GraphDef', to solve this error I have changed the tf.GraphDef to tf.compact.v1.GraphDef

After that I got error as: NHWC on device type CPU

when I check as:

print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU')))
I got: Num GPUs Available: 0

Cuda Version → 10.0
Cudnn Version → 7.6

But to run with tensorflow 2.4.0 we need :
Cuda Version → 11.0
Cudnn Version → 8.0

hence it cant figure out the GPU device.

But, when I run the same second step with tensorflow== 1.15.0, It give me another error as:
tensorflow.python.framework.errors_impl.NotFoundError: /content/hifi3dface/third_party/kernels/rasterize_triangles_kernel.so
When I manually tried to put the .so file in directory it gave me "Invalid ELF error"

My question is how could same script run with two different tensorflow, cuda, cudnn version.
Can you please list final,
Python Version →
Cuda Version →
Tensorflow Version →
Cudnn Version →

Thanks