DSAverse

Initializing Sorting Algorithms...

Sorting

Trees

Graphs

Preparing interactive visualizations...

Cheatsheet Review Compare Contact

Deep Learning Architectures

Sliding the kernel…

DSAverse

Making data structures, algorithms, core computer science, mathematics and AI accessible through interactive visualizations.

Practice

Review
Compare
Complexity Cheatsheet
Home
Contact

Data Structures & Algorithms

Basics
Sorting
Searching
Trees
Graph Algorithms
Dynamic Programming
Indexing

CS Core

Operating Systems
Computer Networks
Database Systems
Under the Hood

Math

Combinatorics
Probability and Statistics
Vectors
Linear Algebra
Differential Calculus
Integral Calculus

AI & ML

Machine Learning
Neural Networks
Deep Learning Architectures
Transformers

© 2025 DSAverse. Built withfor better learning.

Back to Deep Learning Architectures

Sequence-to-Sequence

One network reads a sentence, another writes its translation, and between them sits a single vector of 32 numbers. Watch a trained encoder–decoder translate, and see what happens when the sentence no longer fits.

Builds onLSTM and GRU

Trained for this site · GRU encoder–decoder · 9,744 numbers

Loading the trained translators (180 KB)…

Two networks, one message

Encoder: a GRU that reads the source, one word at a time.
Vector: its last state, 32 numbers for any sentence.
Decoder: a GRU started from that vector, writing one word per step and reading back what it wrote.
Training feeds the decoder the right previous word (teacher forcing); translating feeds it its own.

A common misconception

“The encoder's final state is a summary that holds the whole sentence.”

Measuring once the models have loaded…

Check yourself

Question 1 of 3

In a plain encoder–decoder, what does the decoder receive from the source sentence?

See also

Related elsewhere in AI & ML

Why Attention Encoder-Decoder and Cross-Attention Self-Attention

More in this section