1. run random train-test split first run below to save the random splitted train test data(submit job to run this)

    python generate_train_test.py
   

   then run the inference:

    python IN_dataGenerator_random.py IN_training 1 0 --De 20 --Do 24 --hidden 60
   

2. run the conversion from pytorch to onnx, and use onnx to inference set up environemnt:

    module load conda/2021-06-28
    conda activate base
   

   run model:

    python pytorch2onnx.py
    python pytorch_onnx_inference.py
   

   also include a notebook contain overall process onnx.ipynb

3. run the conversion from onnx to tensorrt, and use tensorrt to inference

    module load conda/2021-11-30
    conda activate base
    python tensorrt_inference.py
   

4. use singularity container start the container: if you are first time use this container, need to build the instance first. After build the instance, use singularity shell --bind to connect the path of dataset:

        singularity instance start  --nv ./trt_torch_new.sif tensorrt
        singularity shell --bind //grand/RAPINS/ruike/new_hbb/test://home/ruike/merge_IN/notebook_code/test_hbb //home/ruike/trt_torch_new.sif ls //home/ruike/merge_IN/notebook_code/test_hbb
        pip install -U scikit-learn
        ...
   

5. throughput & gpu throughput plot the plots are saved in throughput_gpu_use.ipynb file