A Command line utility to modify serialized PyTorch model states.
pip install ptmod
In the context of this text, a model file should be a serialized state_dict
object. See for mote detail: https://pytorch.org/docs/stable/notes/serialization.html
# e.g.
import torch
class Block(torch.nn.Module):
def __init__(self):
super().__init__()
self.layer1 = torch.nn.Linear(10, 10)
self.layer2 = torch.nn.Linear(10, 10)
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
self.block1 = Block()
self.block2 = Block()
model = Model()
torch.save(model.state_dict(), "model.pth")
ptmod-ls model.pth
# OR
ptmod "ls model.pth"
% ptmod "ls model.pth"
block1.layer1.weight
block1.layer1.bias
block1.layer2.weight
block1.layer2.bias
block2.layer1.weight
block2.layer1.bias
block2.layer2.weight
block2.layer2.bias
% ptmod "ls -l model.pth"
block1.layer1.weight (10, 10)
block1.layer1.bias (10,)
block1.layer2.weight (10, 10)
block1.layer2.bias (10,)
block2.layer1.weight (10, 10)
block2.layer1.bias (10,)
block2.layer2.weight (10, 10)
block2.layer2.bias (10,)
ptmod-cp model.pth:key out.pth:key2
# OR
ptmod "cp model.pth:key out.pth:key2"
% ptmod "cp model.pth:block1 out.pth" "ls out.pth"
layer1.weight
layer1.bias
layer2.weight
layer2.bias
% ptmod "cp model.pth:block1 out.pth:foo" "ls out.pth"
foo.layer1.weight
foo.layer1.bias
foo.layer2.weight
foo.layer2.bias
ptmod-rm model.pth:key
# OR
ptmod "rm model.pth:key"
% ptmod "cp model.pth out.pth" "rm out.pth:block1" "ls out.pth"
block2.layer1.weight
block2.layer1.bias
block2.layer2.weight
block2.layer2.bias
% ptmod "cp model.pth out.pth" "rm out.pth:block2.layer2" "ls out.pth"
block2.layer1.weight
block2.layer1.bias
block1.layer1.weight
block1.layer1.bias
block1.layer2.weight
block1.layer2.bias
ptmod-average out.pth model1.pth model2.pth ...
# OR
ptmod "average out.pth model1.pth model2.pth ...
% ptmod "average out.pth model1.pth model2.pth"
% ptmod "sum out.pth model1.pth model2.pth"