AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Context-level Language Modeling by Learning Predictive Context Embeddings

beiya dai, Yuliang Liu, Yunchong Song, Daozheng Xue, Qipeng Guo, Kai Chen, Xinbing Wang, Bowen Zhou, Zhouhan Lin

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Language model pretraining and induction

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 next-context prediction (ContextLM) to GPT-2 BaseGPT-2 Base · OpenWebText-trained avg perplexityavg perplexity: 110.92 -> 87.19Improved
add next-context prediction (ContextLM) to GPT-2 XLGPT-2 XL · OpenWebText-trained avg perplexityavg perplexity: 31.54 -> 25.93Improved
add two standard Transformer layers (parameter-matched) to GPT-2 XL (ablation)GPT-2 XL · OpenWebText-trained avg perplexityavg perplexity: 31.54 -> 31.57No clear effect
add two standard Transformer layers (parameter-matched) to GPT-2 Base (ablation)GPT-2 Base · OpenWebText-trained avg perplexityavg perplexity: 110.92 -> 107.77Improved
add ContextLM to GPT-2 Large relative to parameter-matched baseline (ablation)GPT-2 Large · 9 downstream benchmarks avg accuracy (zero-shot)avg accuracy: 42.4 -> 43.9Improved
add ContextLM to Pythia-70MPythia-70M · Pythia avg perplexityavg perplexity: 297.72 -> 142.71Improved
add ContextLM to Pythia-1.4BPythia-1.4B · Pythia avg perplexityavg perplexity: 9.74 -> 9.54Improved
add ContextLM to Pythia-70MPythia-70M · 9 downstream benchmarks avg accuracy (zero-shot)avg accuracy: 34.6 -> 38.2Improved
add ContextLM to Pythia-1.4BPythia-1.4B · 9 downstream benchmarks avg accuracy (zero-shot)avg accuracy: 54.1 -> 54.4Improved
add ContextLM to Pythia-70M, five-shotPythia-70M · 9 downstream benchmarks avg accuracy (five-shot)avg accuracy: 32.6 -> 36.7Improved
fine-tune ContextLM-Pythia instead of Pythia on AlpacaPythia-1.4B · AlpacaEval 2.0win rate (%): 45.52 -> 54.48Improved
add ContextLM to Pythia-1.4B for long contextPythia-1.4B · LongBench 0-4koverall avg: 20.21 -> 23.32Improved

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