Encoder-Decoder
Definition pending
We've cataloged "Encoder-Decoder" but haven't written a full definition yet. Definitions are hand-curated rather than auto-generated, so it takes time to cover every term.
Want this one prioritized? Email us and we'll bump it.
Practical example
Encoder-decoder architecture: encoder reads input, decoder generates output. Used in T5, BART, and most translation models. The operational advantage: input and output can be different lengths, modalities, or even languages. The disadvantage: two separate stacks = more parameters, more VRAM than a pure decoder of comparable quality.
Workflow example
Deploy encoder-decoder for translation: load a T5 or NLLB model. These are ~3B params (vs 7B for comparable decoder-only) but produce higher-quality translations because the encoder deeply understands the source. Benchmark: compare translation BLEU scores for encoder-decoder (T5-large) vs decoder-only (Llama 3 8B with few-shot prompt) on your language pair. The specialized model usually wins.