TorchScript for production in PyTorch - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to convert a PyTorch model to TorchScript using tracing.

PyTorch

import torch
import torchvision.models as models

model = models.resnet18(pretrained=True)
model.eval()
example_input = torch.randn(1, 3, 224, 224)
scripted_model = torch.jit.[1](model, example_input)

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

Atrace

Bscript

Ccompile

Doptimize

Attempts:

3 left

💡 Hint

Common Mistakes

Using torch.jit.script instead of torch.jit.trace for models with fixed control flow.

Trying to call compile which is not a torch.jit method.

Tracing records the operations of the model with example inputs to create a TorchScript version.

2fill in blank

medium

Complete the code to save the TorchScript model to a file.

PyTorch

scripted_model = torch.jit.trace(model, torch.randn(1, 3, 224, 224))
scripted_model.[1]('resnet18_scripted.pt')

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

Asave

Bwrite

Cexport

Ddump

Attempts:

3 left

💡 Hint

Common Mistakes

Using export or write which are not TorchScript methods.

Trying to use dump which is not a valid method here.

The save() method writes the scripted model to disk for later use.

3fill in blank

hard

Fix the error in loading a TorchScript model from a file.

PyTorch

import torch

scripted_model = torch.jit.[1]('resnet18_scripted.pt')

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

Aread

Bload

Copen

Dimport

Attempts:

3 left

💡 Hint

Common Mistakes

Using open or read which are Python file functions but not for TorchScript models.

Using import which is not a torch.jit method.

torch.jit.load() loads a saved TorchScript model from disk.

4fill in blank

hard

Fill both blanks to define a TorchScript function and save it.

PyTorch

import torch

@torch.jit.[1]
def add_tensors(x, y):
    return x + y

scripted_fn = add_tensors
scripted_fn.[2]('add_tensors.pt')

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

Ascript

Bsave

Ctrace

Dexport

Attempts:

3 left

💡 Hint

Common Mistakes

Using trace decorator which does not exist.

Using export instead of save to write the file.

Use @torch.jit.script to compile the function and save() to write it to a file.

5fill in blank

hard

Fill all three blanks to create a scripted model, run inference, and print the output shape.

PyTorch

import torch
import torchvision.models as models

model = models.mobilenet_v2(pretrained=True)
model.eval()
example_input = torch.randn(1, 3, 224, 224)
scripted_model = torch.jit.[1](model, example_input)
output = scripted_model([2])
print(output.[3])

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

Atrace

Bexample_input

Cshape

Dscript

Attempts:

3 left

💡 Hint

Common Mistakes

Using script instead of trace for this model.

Passing wrong input variable to the model.

Printing output instead of output.shape.

Trace the model, pass example_input to get output, then print output.shape to see the tensor size.