Add new file

model_parser.py

+import torch
+import sys
+import argparse
+import numpy as np
+
+
+#This is just some setup for getting the name of the model if you excecute this file
+parser = argparse.ArgumentParser(description='Get the name of the model')
+parser.add_argument('m_path', metavar='model_path', type=str, help='enter the name of the model')
+args = parser.parse_args()
+
+#The function looks for the model with the name given to it and returns every weight matrix as a numpy array
+#if you already know which matrix you are interessted in you can give it to the function as a string and only this matrix will be returned
+def model_parser(path,layer=None):
+        model=torch.load(path,map_location=torch.device('cpu'))
+        model_dict=model.state_dict()
+        for name in model_dict:
+                model_dict[name]=np.array(model_dict[name])
+        if layer is not None:
+                return model_dict[layer]
+        else:
+                return model_dict
+
+
+
+if __name__ == "__main__":
+        x=model_parser(args.m_path)
+        for name in x:
+                print(name)
+                print("shape: "+str(x[name].shape))
+