AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Cascaded Text Generation with Markov Transformers

Yuntian Deng, Alexander Rush

NeurIPS 2020 · Architectures and efficiency › Attention and transformers · approach: Efficient language transformers · 12 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add knowledge distillation (train on autoregressive teacher outputs) (ablation)Cascaded Generation (K=64, iters=2) · WMT14 En-De testBLEU: 23.09 -> 26.52Improved
relax length constraint to multiple lengths (DeltaL=3 vs 0) (ablation)Cascade (K=64, iters=5) · IWSLT14 De-En valBLEU: 33.45 -> 35.03Improved
use max-marginals instead of n-gram scores for pruning (ablation)Cascaded decoding · WMT14 En-De valBLEU: 28.42 -> 29.24Improved
use cascaded decoding instead of serial beam search without distillation and no length constraint (ablation)Markov Transformer · IWSLT14 De-En valBLEU: 35.07 -> 33.45Hurt
use cascaded decoding instead of beam search under hard length constraint (ablation)Markov Transformer · IWSLT14 De-En valBLEU: 33.33 -> 33.45Improved
increase iterations (higher Markov order) from 2 to 5 (ablation)Cascaded Generation (K=64) · WMT14 En-De testBLEU: 26.52 -> 26.92Improved
use 4-order Markov transformer vs fully autoregressive transformer with beam 5Markov Transformer · IWSLT14 De-En valBLEU: 35.63 -> 35.07Hurt
add more iterations to reduce repetitions (ablation)Cascaded decoding · WMT14 En-De valratio of unique n-gramsImproved
parallelize decoder across 4 GPUs instead of 1Cascaded decoding K=64 iters=5 · WMT14 En-De testspeedup: 1.68x -> 2.79xImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy