The full transformer architecture consists of an encoder component and a decoder component. For models to specialize in accomplishing certain tasks and/or to best handle specific types of input/output data, some language models only use the encoder component, only use the decoder component, or use both the encoder and decoder (i.e., use the full transformer architecture).

Question

BlackTom AI · Accepted Answer

The Transformer architecture includes distinct encoder and decoder stacks that can be used together (encoder-decoder) or separately. This modularity lets model builders select only the encoder, only the decoder, or both, depending on the task and data.

1) Relevant knowledge points and concepts
- Transformer building blocks: the encoder processes input sequences to produce contextualized representations; the decoder generates output sequences autoregressively, attending to encoder outputs when used in full sequence-to-sequence setups.
- Model specializations: encoder-only models (e.g., BERT) are optimized for understanding tasks; decoder-only models (e.g., GPT series) are optimized for text generation; encoder-decoder (e.g., T5, BART) are optimized for sequence-to-sequence tasks like translation, summarization, and many conditional generation tasks.

2) Step-by-step reasoning
- The full Transformer contains both encoder and decoder modules.
- If a task requires only representation/understanding (classification, QA encoding, embedding), the decoder is unnecessary—an encoder-only model suffices and is more efficient.
- If a task is unconditional or causal generation, a decoder-only design with causal masking is appropriate.
- If a task maps input sequences to output sequences (translation), both encoder and decoder are used so the decoder can condition on encoder outputs.

3) Calculations
- No numerical calculations are required; reasoning is architectural and conceptual.

4) Why the correct answer is correct
- The statement matches architectural practice and observed model designs: different families of language models explicitly adopt encoder-only, decoder-only, or encoder-decoder topologies to specialize for specific tasks and input/output formats.

5) Why the other option is incorrect
- “False” would deny the modular reality of Transformer implementations. That contradicts widespread models and design patterns (BERT, GPT, T5, BART), so it is incorrect.

6) Conclusion
- Therefore the correct answer is True, which matches the provided answer.

类似问题

Which of the following best describes a key property of Transformer architecture used in LLMs?

The transformer architecture only consists of a single encoder component and single decoder component.

What is the main architectural innovation in ChatGPT that allows it to handle complex language tasks?

A team of data scientists is experiencing poor performance with their Transformer model. After inspecting their implementation, you find several suspicious design choices. Which one would you identify as the MOST problematic based on the architectural principles discussed in the lecture?

What operations are part of a standard Transformer block? (Select all that apply.)

Is the following statement true or false? In the products of methylation-hydrolysis, every -OH group corresponds to the position of a glycosidic bond in the starting polysaccharide.

Which of the follwoing structures represents amylopectin?

Which of the following is true about an aldopentose?I. It is a monosaccharide.II. It contains a CHO groupIII. It is a disaccharide.IV. It is an oligosaccharide.

更多留学生实用工具

智能学习助手

风格化写作助手

论文查重助手

文献引用助手

课堂转译助手

课堂笔记助手

Quiz搜索助手

学校历年真题

智能刷题助手

智能匹配练习

希望你的学习变得更简单