Skip to main content

A Command line utility to modify serialized PyTorch model states.

Project description

ptmod: Modify PyTorch model

A Command line utility to modify serialized PyTorch model states.

The definition of pytorch model file

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")

Show the contents in a model file

Usage

ptmod-ls model.pth
# OR
ptmod "ls model.pth"

Example

% 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,)

Copy parameters

Usage

ptmod-cp model.pth:key out.pth:key2
# OR
ptmod "cp model.pth:key out.pth:key2"

Example

% 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

Remove parameters

Usage

ptmod-rm model.pth:key
# OR
ptmod "rm model.pth:key"

Example

% 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

Sum/Average parameters

Usage

ptmod-average out.pth model1.pth model2.pth ...
# OR
ptmod "average out.pth model1.pth model2.pth ...

Example

% ptmod "average out.pth model1.pth model2.pth"
% ptmod "sum out.pth model1.pth model2.pth"

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ptmod-0.0.0.tar.gz (3.4 kB view hashes)

Uploaded Source

Built Distribution

ptmod-0.0.0-py3-none-any.whl (4.4 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page