AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Learning to Break the Loop: Analyzing and Mitigating Repetitions for Neural Text Generation

Jin Xu, Xiaojiang Liu, Jianhao Yan, Deng Cai, Huayang Li, Jian Li

NeurIPS 2022 · Large language models › Language, safety and interpretability · approach: Text generation and evaluation metrics · 21 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
fine-tune with DITTO pseudo-repetition penalization instead of plain MLE16-layer Transformer on Wikitext-103 · Wikitext-103 test, greedy decodingMAUVE: 0.34 -> 0.77Improved
fine-tune with DITTO instead of MLE16-layer Transformer on Wikitext-103 · Wikitext-103 test, greedy decodingPerplexity: 25.68 -> 24.33Improved
fine-tune with DITTO instead of MLE16-layer Transformer on Wikitext-103 · Wikitext-103 test, greedy decodingRepetition-Sen: 14.50 -> 2.85Improved
DITTO versus UL-token+seq on repetition rate16-layer Transformer on Wikitext-103 · Wikitext-103 test, greedy decodingRepetition-4: 7.60 -> 22.00Hurt
fine-tune with DITTO instead of MLE under top-k and nucleus sampling16-layer Transformer on Wikitext-103 · Wikitext-103 test, nucleus samplingMAUVE: 0.94 -> 0.96Improved
fine-tune BART-large with DITTO instead of MLEBART-large · CNN/DailyMailROUGE-1: 44.11 -> 44.41Improved
replace DITTO loss with MSE loss (ablation)DITTO · Wikitext-103 test, greedy decodingMAUVE: 0.77 -> 0.73Hurt
replace DITTO loss with margin (hinge) loss (ablation)DITTO · Wikitext-103 test, greedy decodingMAUVE: 0.77 -> 0.66Hurt
repeat two-sentence groups instead of one sentence for pseudo data (ablation)DITTO · Wikitext-103 test, greedy decodingMAUVE: 0.77 -> 0.77No clear effect
shorten pseudo data to 150 tokens (DITTO-short) (ablation)DITTO · Wikitext-103 self-reinforcement (TP/IP/WR)TP, IP, WRHurt
mix pseudo and real data at gamma=0.5 (versus other ratios) (ablation)DITTO · Wikitext-103 and CNN/DailyMailgeneration qualityImproved
stronger penalty lambda=0.5 rather than 0.9 (ablation)DITTO · Wikitext-103 vs CNN/DailyMailgeneration qualityMixed

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