Multi-head attention in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the main idea behind multi-head attention?

Multi-head attention splits the attention mechanism into several smaller parts (heads) that run in parallel. Each head learns different relationships in the data, and their results are combined to capture richer information.

beginner

In multi-head attention, what are the Query, Key, and Value?

Query, Key, and Value are three sets of vectors derived from the input data. The Query is what we want to find information about, the Key helps match the Query, and the Value holds the actual information to be gathered.

intermediate

Why do we use multiple heads instead of one in attention?

Using multiple heads allows the model to focus on different parts or aspects of the input simultaneously. This helps the model understand complex patterns better than a single attention head.

intermediate

What is the shape of the output from a multi-head attention layer in PyTorch?

The output shape is usually (sequence_length, batch_size, embedding_dim), where embedding_dim is the original input dimension after combining all heads.

intermediate

How is the scaled dot-product attention computed inside each head?

Scaled dot-product attention is computed by taking the dot product of Query and Key, dividing by the square root of the key dimension to scale, applying softmax to get weights, and then multiplying by the Value vectors.

What does each head in multi-head attention learn?

A Different relationships or features from the input

B The same information repeatedly

C Only the first token of the input

D Random noise

Each head focuses on different parts or aspects of the input to capture diverse information.

What is the purpose of scaling the dot product in scaled dot-product attention?

A To normalize the output to zero mean

B To prevent large values that can make softmax gradients too small

C To increase the dot product values

D To reduce the number of heads

Scaling by the square root of the key dimension keeps the dot products in a range that avoids very small gradients after softmax.

In PyTorch, which class implements multi-head attention?

A torch.nn.AttentionHead

B torch.nn.AttentionLayer

C torch.nn.MultiHeadLayer

D torch.nn.MultiheadAttention

PyTorch provides torch.nn.MultiheadAttention for multi-head attention implementation.

What is combined after all attention heads compute their outputs?

A The outputs are concatenated and projected back to the original dimension

B Only the first head's output is used

C The outputs are averaged without projection

D The outputs are discarded

Outputs from all heads are concatenated and then passed through a linear layer to match the original embedding size.

Which of these is NOT a component of multi-head attention?

A Query

B Value

C Bias

D Key

Query, Key, and Value are core components; bias is not a fundamental part of multi-head attention.

Explain how multi-head attention works and why it is useful in simple terms.

Think about how looking at something from different angles helps understand it better.

Describe the role of Query, Key, and Value in the attention mechanism.

Imagine searching for a book in a library using a catalog.

Practice

1. What is the main purpose of multi-head attention in PyTorch's MultiheadAttention layer?

easy

A. To reduce the size of the input data

B. To perform simple addition of input vectors

C. To increase the number of layers in the model

D. To allow the model to focus on different parts of the input simultaneously

Solution

1. Step 1: Understand the concept of multi-head attention

   Multi-head attention splits the input into multiple parts (heads) to focus on different information simultaneously.

2. Step 2: Relate to PyTorch's MultiheadAttention layer

   This layer implements this idea to improve model understanding by attending to various input features at once.

3. Final Answer:

   To allow the model to focus on different parts of the input simultaneously -> Option D

4. Quick Check:

   Multi-head attention = focus on multiple input parts [OK]

Hint: Multi-head means multiple focus points at once [OK]

Common Mistakes:

• Thinking it reduces input size
• Confusing it with adding layers
• Assuming it just adds vectors

2. Which of the following is the correct way to create a multi-head attention layer with 8 heads and embedding dimension 64 in PyTorch?

easy

A. torch.nn.MultiheadAttention(8, 64)

B. torch.nn.MultiheadAttention(embed_dim=64, num_heads=8)

C. torch.nn.MultiheadAttention(num_heads=64, embed_dim=8)

D. torch.nn.MultiheadAttention(64)

Solution

1. Step 1: Check PyTorch MultiheadAttention parameter order

   The constructor requires embed_dim first, then num_heads as keyword arguments.

2. Step 2: Verify correct syntax

   torch.nn.MultiheadAttention(embed_dim=64, num_heads=8) correctly uses embed_dim=64 and num_heads=8. torch.nn.MultiheadAttention(num_heads=64, embed_dim=8) swaps parameters incorrectly, A and D use positional arguments wrongly.

3. Final Answer:

   torch.nn.MultiheadAttention(embed_dim=64, num_heads=8) -> Option B

4. Quick Check:

   embed_dim first, num_heads second [OK]

Hint: Remember: embed_dim first, num_heads second in constructor [OK]

Common Mistakes:

• Swapping embed_dim and num_heads
• Using positional args incorrectly
• Missing one of the required parameters

3. Given the following PyTorch code, what is the shape of the output tensor attn_output?

import torch
mha = torch.nn.MultiheadAttention(embed_dim=16, num_heads=4)
query = torch.rand(5, 3, 16)  # (sequence_length, batch_size, embed_dim)
key = torch.rand(5, 3, 16)
value = torch.rand(5, 3, 16)
attn_output, attn_weights = mha(query, key, value)

medium

A. (5, 3, 16)

B. (3, 5, 16)

C. (5, 16, 3)

D. (3, 16, 5)

Solution

1. Step 1: Understand input shapes for MultiheadAttention

   Inputs are (sequence_length, batch_size, embed_dim). Here, query shape is (5, 3, 16).

2. Step 2: Check output shape of attn_output

   Output shape matches query shape: (5, 3, 16).

3. Final Answer:

   (5, 3, 16) -> Option A

4. Quick Check:

   Output shape = query shape = (5, 3, 16) [OK]

Hint: Output shape matches query input shape [OK]

Common Mistakes:

• Mixing batch and sequence dimensions
• Confusing embed_dim position
• Assuming output shape changes with heads

4. Identify the error in this PyTorch multi-head attention code snippet:

import torch
mha = torch.nn.MultiheadAttention(embed_dim=32, num_heads=4)
query = torch.rand(10, 2, 32)
key = torch.rand(10, 2, 16)
value = torch.rand(10, 2, 32)
output, weights = mha(query, key, value)

medium

A. The embed_dim of key does not match embed_dim of query and value

B. The batch size is incorrect

C. The sequence length is too large

D. The number of heads is too high

Solution

1. Step 1: Check embed_dim consistency

   MultiheadAttention requires query, key, and value to have the same embedding dimension. Here, key has embed_dim 16, while query and value have 32.

2. Step 2: Identify the error cause

   This mismatch causes a runtime error because the attention mechanism cannot align different embedding sizes.

3. Final Answer:

   The embed_dim of key does not match embed_dim of query and value -> Option A

4. Quick Check:

   Embed dims must match for query, key, value [OK]

Hint: Check all embed_dims match before calling attention [OK]

Common Mistakes:

• Ignoring embed_dim mismatch
• Assuming batch size causes error
• Thinking sequence length is the problem

5. You want to apply multi-head attention to a batch of sentences with varying lengths. Which PyTorch feature helps you handle padding tokens correctly during attention computation?

hard

A. Increasing the number of heads to cover padding

B. Setting batch_first=True in MultiheadAttention

C. Using the key_padding_mask argument in MultiheadAttention

D. Manually removing padding tokens before attention

Solution

1. Step 1: Understand padding in variable-length sequences

   Sentences have different lengths, so padding tokens are added to make batch tensors uniform.

2. Step 2: Use key_padding_mask to ignore padding

   PyTorch's MultiheadAttention accepts key_padding_mask to mask out padded positions during attention, preventing them from affecting results.

3. Final Answer:

   Using the key_padding_mask argument in MultiheadAttention -> Option C

4. Quick Check:

   Mask padding tokens with key_padding_mask [OK]

Hint: Use key_padding_mask to ignore padded tokens [OK]

Common Mistakes:

• Removing padding manually (inefficient)
• Confusing batch_first with padding handling
• Increasing heads does not fix padding