requires_grad flag in PyTorch - Model Metrics & Evaluation

Metrics & Evaluation - requires_grad flag

Which metric matters for the requires_grad flag and WHY

The requires_grad flag in PyTorch controls whether a tensor tracks operations for gradient calculation. It does not directly affect model accuracy or loss but impacts training effectiveness. Metrics like loss decrease and gradient norms matter because if requires_grad is False, gradients won't be computed, and the model won't learn. So, the key metric is whether gradients are correctly computed and used to update model weights.

Confusion matrix or equivalent visualization

Since requires_grad is about gradient tracking, a confusion matrix is not applicable. Instead, consider this simple illustration of gradient flow:

    Tensor A (requires_grad=True) --> Operations --> Gradient computed
    Tensor B (requires_grad=False) --> Operations --> No gradient computed
    

If requires_grad=False, gradients are None, so no learning happens.

Precision vs Recall tradeoff analogy for requires_grad

Think of requires_grad=True as turning on a learning mode. If you forget to set it, your model won't learn (like a student not paying attention). Setting it True everywhere might slow training or use more memory, like trying to learn everything at once.

Tradeoff:

• True: Model learns but uses more memory and compute.
• False: Saves memory but no learning happens for those tensors.

Use requires_grad=True for parameters you want to train, and False for fixed parts (like pretrained layers you don't want to change).

What "good" vs "bad" metric values look like for requires_grad

Good:

• Gradients are computed for all trainable parameters (non-zero gradient norms).
• Loss decreases steadily during training.
• Model weights update correctly.

Bad:

• Gradients are zero or None for parameters that should learn.
• Loss stays flat or does not improve.
• Model weights do not change after training steps.

Common pitfalls with requires_grad flag

• Forgetting to set requires_grad=True on model parameters, so no learning happens.
• Setting requires_grad=True on tensors that should remain fixed, wasting memory and compute.
• Mixing tensors with different requires_grad flags causing unexpected gradient flow.
• Not detaching tensors when needed, causing memory leaks.
• Assuming accuracy or loss alone shows requires_grad issues; always check gradients.

Self-check question

Your model has 98% accuracy but the gradients for some parameters are always zero because their requires_grad flag is False. Is this model good for training? Why or why not?

Answer: No, it is not good for training those parameters. Even if accuracy looks high, parameters with requires_grad=False won't update because no gradients are computed. This means the model can't improve those parts, which may limit learning or cause poor generalization.

Key Result

The requires_grad flag controls gradient computation; correct setting ensures gradients exist and loss decreases during training.

Practice

1. What does setting requires_grad=True on a PyTorch tensor do?

easy

A. It tells PyTorch to track operations on the tensor for gradient calculation.

B. It freezes the tensor so it won't change during training.

C. It converts the tensor to a NumPy array.

D. It deletes the tensor from memory.

Solution

1. Step 1: Understand the purpose of requires_grad

   The flag requires_grad=True tells PyTorch to remember all operations on the tensor so it can compute gradients later.

2. Step 2: Identify what happens when requires_grad is True

   This tracking is essential for training models because gradients are used to update weights.

3. Final Answer:

   It tells PyTorch to track operations on the tensor for gradient calculation. -> Option A

4. Quick Check:

   requires_grad=True means track gradients [OK]

Hint: True means track gradients for learning, False means no tracking [OK]

Common Mistakes:

• Thinking requires_grad freezes the tensor
• Confusing requires_grad with tensor conversion
• Assuming it deletes the tensor

2. Which of the following is the correct way to create a tensor with gradient tracking enabled?

easy

A. torch.tensor([1.0, 2.0, 3.0], grad=True)

B. torch.tensor([1.0, 2.0, 3.0], requires_grad=False)

C. torch.tensor([1.0, 2.0, 3.0], requires_grad=True)

D. torch.tensor([1.0, 2.0, 3.0], track_grad=True)

Solution

1. Step 1: Recall the correct argument name

   The correct argument to enable gradient tracking is requires_grad=True.

2. Step 2: Check each option

   torch.tensor([1.0, 2.0, 3.0], requires_grad=True) uses requires_grad=True, which is correct. Options C and D use invalid argument names, and B disables gradient tracking.

3. Final Answer:

   torch.tensor([1.0, 2.0, 3.0], requires_grad=True) -> Option C

4. Quick Check:

   Use requires_grad=True to track gradients [OK]

Hint: Use requires_grad=True exactly to enable gradients [OK]

Common Mistakes:

• Using incorrect argument names like grad or track_grad
• Setting requires_grad to False by mistake
• Forgetting to set requires_grad at tensor creation

3. What will be the output of this code snippet?

import torch
x = torch.tensor([2.0, 3.0], requires_grad=True)
y = x * 2
y.backward(torch.tensor([1.0, 1.0]))
print(x.grad)

medium

A. tensor([2., 2.])

B. tensor([1., 1.])

C. tensor([4., 6.])

D. None

Solution

1. Step 1: Understand the operations and gradient flow

   y = x * 2 means each element in x is multiplied by 2. The gradient of y with respect to x is 2 for each element.

2. Step 2: Apply backward with gradient vector

   Calling y.backward with tensor([1.0, 1.0]) sums gradients weighted by 1. So x.grad will be tensor([2., 2.]).

3. Final Answer:

   tensor([2., 2.]) -> Option A

4. Quick Check:

   Gradient of 2*x is 2 [OK]

Hint: Gradient of x*2 is 2, so x.grad = [2, 2] [OK]

Common Mistakes:

• Expecting gradient to be the original tensor values
• Confusing output with input tensor
• Forgetting to call backward()

4. Identify the error in this code snippet related to gradient tracking:

import torch
x = torch.tensor([1.0, 2.0, 3.0])
y = x * 3
y.backward()

medium

A. x was not created with requires_grad=True, so backward() will fail.

B. There is no error; the code runs fine.

C. The backward() call is missing a gradient argument.

D. The tensor y is not a scalar, so backward() cannot be called.

Solution

1. Step 1: Check requires_grad flag on x

   x was created without requires_grad=True, so it does not track gradients. This means y also does not track gradients, causing backward() to fail.

2. Step 2: Understand backward() requirements

   The error occurs because tensors need grad_fn (from requires_grad=True) to call backward(). Additionally, non-scalar tensors need a gradient argument, so calling backward() without it on a vector tensor causes an error.

3. Final Answer:

   The tensor y is not a scalar, so backward() cannot be called. -> Option D

4. Quick Check:

   Non-scalar tensors require gradient argument in backward() [OK]

Hint: Non-scalar tensors need gradient argument in backward() [OK]

Common Mistakes:

• Forgetting requires_grad=True on tensors before backward()
• Ignoring requires_grad flag importance
• Not providing gradient argument for vector tensors

5. You want to freeze the first layer of a neural network during training to save memory. Which code snippet correctly disables gradient tracking for that layer's parameters?

hard

A. model.layer1.grad = None

B. for param in model.layer1.parameters():
param.requires_grad = False

C. model.layer1.requires_grad = False

D. for param in model.layer1.parameters():
param.requires_grad = True

Solution

1. Step 1: Understand how to freeze layers

   Freezing means stopping gradient calculation for parameters by setting their requires_grad to False.

2. Step 2: Check correct way to set requires_grad

   Parameters are tensors inside the layer. We must set param.requires_grad = False for each parameter. for param in model.layer1.parameters(): param.requires_grad = False does this correctly.

3. Final Answer:

   for param in model.layer1.parameters(): param.requires_grad = False -> Option B

4. Quick Check:

   Set requires_grad=False on parameters to freeze [OK]

Hint: Freeze by setting requires_grad=False on each parameter [OK]

Common Mistakes:

• Trying to set requires_grad on the whole layer object
• Setting requires_grad=True instead of False
• Confusing grad attribute with requires_grad