Model packaging (.mar files) in PyTorch - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to save a PyTorch model's state dictionary.

PyTorch

torch.save(model[1], 'model.pth')

Drag options to blanks, or click blank then click option'

A.state_dict()

B.parameters()

C.weights

D.get_state()

Attempts:

3 left

💡 Hint

Common Mistakes

Using model.parameters() instead of model.state_dict()

Trying to save the entire model object directly

We save the model's state dictionary using model.state_dict() to capture the learned parameters.

2fill in blank

medium

Complete the command to create a .mar file using torch-model-archiver.

PyTorch

torch-model-archiver --model-name my_model --version 1.0 --serialized-file [1] --handler handler.py --export-path model_store

Drag options to blanks, or click blank then click option'

Amodel.pth

Bmodel.pt

Cmodel.pkl

Dmodel.json

Attempts:

3 left

💡 Hint

Common Mistakes

Using wrong file extensions like .json or .pkl

Confusing the handler file with the serialized file

The serialized file is the saved model weights file, typically named model.pth.

3fill in blank

hard

Fix the error in loading a model from a .mar file using TorchServe.

PyTorch

torchserve --start --model-store model_store --models my_model=[1]

Drag options to blanks, or click blank then click option'

Amy_model.py

Bmy_model-1.0.mar

Chandler.py

Dmodel.pth

Attempts:

3 left

💡 Hint

Common Mistakes

Using the raw weights file instead of the .mar file

Using the handler or source code file name

When starting TorchServe, the model must be referenced by its .mar file name.

4fill in blank

hard

Fill both blanks to define a custom handler class for TorchServe.

PyTorch

from ts.torch_handler.base_handler import {{BLANK_1 }}

class CustomHandler({{BLANK_2}}):
    def preprocess(self, data):
        pass

Drag options to blanks, or click blank then click option'

ABaseHandler

BTorchHandler

CBaseModel

DTorchBase

Attempts:

3 left

💡 Hint

Common Mistakes

Using incorrect base class names like TorchHandler or BaseModel

The custom handler should inherit from BaseHandler to implement TorchServe methods.

5fill in blank

hard

Fill all three blanks to create a dictionary for model metadata in the .mar packaging process.

PyTorch

model_info = {
    'model_name': [1],
    'model_version': [2],
    'handler': [3]
}

Drag options to blanks, or click blank then click option'

A'my_model'

B'1.0'

C'handler.py'

D'model.pth'

Attempts:

3 left

💡 Hint

Common Mistakes

Using the model weights file name instead of the handler file

Not using quotes around string values

The dictionary keys hold the model name, version, and handler script file name as strings.