Seq2Seq with Cross Attention

Step-by-step architectural trace. Features auto-play, real word translation mapping, and attention heatmaps.

Architecture:

Real Words Example

Time: 1 | Ready

1. Alignment

2. Softmax (Weights)

3. Context Vector

4. Concatenation

5. Update State

Source Sequence

Encoder States

Decoder State

Target Sequence

Navigate using the controls or ← → keys.