Sequence-to-Sequence Models Quiz

Answers and explanations appear as you take the quiz.

1. What are the two main components of a sequence-to-sequence model?
   • An encoder and a decoder
   • A tokenizer and a classifier
   • Two identical encoders
   • A filter and a ranker
2. What key problem did the attention mechanism solve in seq2seq models?
   • Slow tokenization
   • The fixed-length vector bottleneck for long inputs
   • Lack of a vocabulary
   • Excessive training data
3. What architecture did the original 2014 seq2seq translation model use?
   • Convolutional networks
   • Decision trees
   • Stacked LSTMs
   • Bare feed-forward layers
4. How does attention build the context for each decoding step?
   • By averaging all input tokens equally
   • By ignoring the encoder entirely
   • By copying the last input token
   • By a weighted sum of encoder states using alignment scores
5. Why can seq2seq outputs differ in length from inputs?
   • The decoder generates tokens until it emits a stop signal
   • Inputs and outputs are always equal length
   • The encoder pads everything
   • Length is fixed in advance
6. Which modern models are popular Transformer-based seq2seq architectures?
   • ResNet and VGG
   • T5 and BART
   • K-means and PCA
   • Word2Vec and GloVe
7. In encoder-decoder attention, which representations are commonly used as keys and values?
   • Only the decoder’s future hidden states
   • The vocabulary file’s byte offsets
   • The encoder’s output representations
   • The optimizer’s learning-rate history
8. What does a bidirectional encoder provide to an attention-based decoder?
   • A fixed one-token output regardless of the input
   • A guarantee that output length equals input length
   • A causal mask over future target tokens
   • Contextual representations of source tokens that can use information from both directions