Question 6Which transformer-based model architecture has the objective of guessing a masked token based on the previous sequence of tokens by building bidirectional representations of the input sequence.1 pointAutoencoderSequence-to-sequenceAutoregressive

In transformer-based language models, what is the significance of the “masking” mechanism ?Question 12Answera. It masks out irrelevant parts of the input sequence to reduce computationb. It allows the model to prioritize certain tokens based on their position in the sequencec.It ensures that rare tokens are given higher attention weightsd.It prevents the model from attending to future tokens during training

What are the encoder and decoder components of a transformer model?The encoder ingests an input sequence and produces a sequence of tokens. The decoder takes in the tokens from the encoder and produces an output sequence.The encoder ingests an input sequence and produces a single hidden state. The decoder takes in the hidden state from the encoder and produces an output sequence.The encoder ingests an input sequence and produces a sequence of hidden states. The decoder takes in the hidden states from the encoder and produces an output sequence.The encoder ingests an input sequence and produces a sequence of images. The decoder takes in the images from the encoder and produces an output sequence.

Question 7Which transformer-based model architecture is well-suited to the task of text translation?1 pointSequence-to-sequenceAutoregressiveAutoencoder8.

What is the main role of the decoder in a Transformer model?Question 14Answera.To generate output tokens based on the final encoder representation.b.To compute attention scores between input and output tokens.c.Learning positional encodings.d.To encode the input sequence.

In a Transformer decoder, what is the purpose of the masked self-attention layer?Question 2Answera.Assign weights to relevant parts of the input sequence.b.None of thesec.Generate a representation of the entire output sequence.d.Allow the model to "attend" to previously generated tokens.