AI Paper Atlas

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

Revisiting Out-of-distribution Robustness in NLP: Benchmarks, Analysis, and LLMs Evaluations

Lifan Yuan, Yangyi Chen, Ganqu Cui, Hongcheng Gao, FangYuan Zou, Xingyi Cheng, Heng Ji, Zhiyuan Liu, Maosong Sun

NeurIPS 2023 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 7 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 FreeLB adversarial trainingvanilla fine-tuning · BOSS DSOOD score: 47.38 -> 47.94Improved
add FreeLB adversarial trainingvanilla fine-tuning · BOSS ANOOD score: 36.19 -> 36.13No clear effect
add model ensemblevanilla fine-tuning · BOSS AZOOD score: 90.94 -> 50.72Hurt
add EDA data augmentationvanilla fine-tuning · BOSS MN/CNOOD score: 89.40 -> 65.57Mixed
add focal lossvanilla fine-tuning · BOSS CoNLLOOD score: 69.10 -> 61.04Hurt
add label smoothingvanilla fine-tuning · BOSS SEOOD score: 49.90 -> 46.35Hurt
use in-context learning instead of zero-shotT5-3B · BOSS SQuADOOD score: 46.64 -> 35.50Hurt
use in-context learning instead of zero-shotDavinci3 · BOSS WNUTOOD score: 32.54 -> 42.10Improved
use in-context learning instead of zero-shotLLaMA-7B · BOSS CCOOD score: 67.72 -> 89.70Improved
use in-context learning instead of zero-shotTurbo · BOSS CoNLLOOD score: 51.36 -> 57.02Improved
use OOD training-split examples (ICL*) instead of ID examplesDavinci3 · BOSS SEOOD score: 68.88 -> 65.14Hurt
use OOD training-split examples (ICL*) instead of ID examplesDavinci3 · BOSS NQAOOD score: 54.79 -> 56.15Improved

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