AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Don’t Stop Pretraining? Make Prompt-based Fine-tuning Powerful Learner

Zhengxiang Shi, Aldo Lipani

NeurIPS 2023 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 6 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 TAPT continued pre-trainingPrompt-based FT (hard) · CoLA, fully supervisedMatthews corr.: 54.7 -> 44.0Hurt
add TAPT continued pre-trainingPrompt-based FT (soft) · QQP, fully supervisedF1: 73.9 -> 79.9Improved
add TAPT continued pre-trainingPrompt-based FT (hard) · MNLI, semi-supervisedaccuracy: 67.3 -> 50.7Hurt
add TAPT continued pre-trainingCLS-based FT · 8 single-sentence tasksaccuracy: 7 out of 8 tasksImproved
add TAPT continued pre-trainingCLS-based FT · 8 sentence-pair tasksaccuracy: 6 out of 8 tasksHurt
add PCP continued pre-trainingPrompt-based FT (hard) · CoLA, semi-supervisedMatthews corr.: 7.2 -> 25.0Improved
add PCP continued pre-trainingPrompt-based FT (soft) · QNLI, semi-supervisedaccuracy: 64.2 -> 84.3Improved
add PCP continued pre-trainingPrompt-based FT · 16 tasks, fully supervisedaccuracy: more than 1.0Improved
add PCP continued pre-trainingPrompt-based FT · 16 tasks, semi-supervisedaccuracy: 6.8Improved
replace semi-supervised methods with PCP-enhanced prompt FTPrompt-based FT (hard) · Amazon Review, 200 labelsMacro-F1: 5.5Improved
compare PCP with FixMatchPrompt-based FT (hard) + PCP · IMDB, 20 labelsMacro-F1: 95.26 vs 92.49Hurt
use only pseudo labels or only template in PCP (ablation)Prompt FT + PCP (soft) · 8 single-sentence tasks, meanmean score: 77.5 -> 73.5 (labels only), 72.2 (template only)Hurt

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