Why do we need Encoder-Decoder Models?
Some tasks only need understanding. Some only need generation. A few need both.
Playground
Which architecture fits the task?
Click a task below. Watch which parts light up - an encoder to understand, a decoder to generate, or both.
Encoder
Decoder
Pick a task to see which parts it needs.
Why both matter
Tasks that read and write
Some of the most useful tasks follow the same pipeline - first read, then understand, then generate.
Read
Understand
Generate
Each of these needs to understand the input and then generate new text. That's why encoder-decoder models exist.
🔑
Key takeaways
- • Some tasks only need understanding.
- • Some tasks only need generation.
- • Some tasks require both.
- • Encoder-decoder models combine both strengths.
Summary
Some AI problems need both understanding and generation. Encoder-decoder models combine these two abilities into one architecture - the idea behind models like T5.