Sequence-to-Sequence Models Mibvunzo

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

1. Ndezvipi zvikamu zviviri zvikuru zvemuenzaniso wenhevedzano-kutevedzana?
   • Encoder uye decoder
   • A tokenizer uye classifier
   • Maencoder maviri akafanana
   • Sefa uye chinzvimbo
2. Nderipi dambudziko rakakosha rakagadziriswa nemaitiro mu seq2seq modhi?
   • Slow tokenization
   • Iyo yakagadziriswa-kureba vector bottleneck yezvekupinza kwenguva refu
   • Kushaikwa kwemazwi
   • Data yekudzidzira yakawandisa
3. Ndechipi chivakwa chakashandiswa neshanduro yekutanga 2014 seq2seq modhi?
   • Convolutional network
   • Sarudzo miti
   • Akaturikidzana LSTMs
   • Bare feed-forward layers
4. Kutarisisa kunovaka sei mamiriro edanho rega rega rekugadzirisa?
   • Nekuita avhareji ma tokeni ese ekupinza zvakaenzana
   • Nekuregeredza encoder zvachose
   • Nekukopa chiratidzo chekupedzisira chekuisa
   • Nehuremu hwakayerwa hweencoder states uchishandisa zvibodzwa zvekuenzanisa
5. Sei seq2seq zvinobuda zvichisiyana pakureba kubva kune zvinopinda?
   • Iyo decoder inogadzira tokeni kusvika yaburitsa chiratidzo chekumira
   • Zvinopinza uye zvinobuda zvinogara zvakaenzana kureba
   • Iyo encoder inobvisa zvese
   • Urefu hunogadziriswa mberi
6. Ndeapi mamodheru emazuva ano anozivikanwa Transformer-based seq2seq architectures?
   • ResNet uye VGG
   • T5 uye BART
   • K-nzira uye PCA
   • Word2Vec uye 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