AI Paper Atlas

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

Calibrating Sequence likelihood Improves Conditional Language Generation

Yao Zhao, Mikhail Khalman, Rishabh Joshi, Shashi Narayan, Mohammad Saleh, Peter J Liu

ICLR 2023 · Large language models › Language, safety and interpretability · approach: Controlled text generation · 38 citations (OpenAlex)

Read the paper

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
Apply sequence likelihood calibration after fine-tuningPEGASUS-XXL · CNN/DailyMail testROUGE-1: 44.31 -> 47.97Improved
Apply sequence likelihood calibration after fine-tuningPEGASUS-XXL · RedditTIFU-long testROUGE-1: 28.73 -> 32.03Improved
Replace token embeddings with contextualized decoder representations for candidate-target similarity (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.51 -> 46.55Improved
Use decoder representations rather than directly optimizing ROUGE as the similarity function (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.47 -> 46.55No clear effect
Use rank loss instead of margin, list-rank, or reward loss (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.73 vs 46.11; 46.62; 46.49Mixed
Add token-level regularization rather than removing it (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.54 -> 46.73Improved
Use KL divergence rather than cross-entropy regularization (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.73 -> 46.80No clear effect
Use beam search rather than diverse beam or nucleus sampling for calibration candidates (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.31 -> 46.50; 46.45 -> 46.50Mixed
Select the fine-tuning checkpoint by perplexity instead of ROUGE (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.66 -> 47.36Improved
Omit length normalization for calibrated decoding (ablation)SLiC · CNN/DailyMail validationROUGE-1: 46.44 -> 46.49No clear effect
Increase evaluation beam candidates for calibrated modelsSLiC · CNN/DailyMail validationROUGE-1: 45.55 -> 46.63Improved
Increase evaluation beam candidates for fine-tuned-only modelsPEGASUS-LARGE fine-tuned · CNN/DailyMail validationROUGE-1: 45.34 -> 44.59Hurt

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