torch convert to TFLite.

https://github.com/google-ai-edge/ai-edge-torch

reference this paper. from https://github.com/google-ai-edge/ai-edge-torch

you need create a new conda envirenment for this project ,point is python >=3.9

i used 3.9 for python.

the deteils are as follows:

python                         3.9

torch                            2.4

protobuf                  3.20.3  

tensorflow                2.17.0 

sympy                     1.8   

opencv-python             4.10.0.82     

numpy                     1.26.4    

keras                     3.5.0   

ai-edge-torch             0.1.1

torch-xla                 2.4.0  

and so on..

remember set the :

export LD_LIBRARY_PATH=<path to Python installation>/lib:$LD_LIBRARY_PATH

<path to Python installation> means the conda's path

ex:/home/yunxi/anaconda3/envs/torch_tf

and add this code for your project:

os.environ["PJRT_DEVICE"]="CPU"

device = "cpu"

for inference add this :

edge_model = ai_edge_torch.convert(model.eval(), (inputs,))

edge_model.export("resnet18.tflite")

print("-------------------tflite save successful !!11!@@@@!!!--------------")