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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add next-context prediction (ContextLM) to GPT-2 Base | GPT-2 Base · OpenWebText-trained avg perplexity | avg perplexity: 110.92 -> 87.19 | Improved |
| add next-context prediction (ContextLM) to GPT-2 XL | GPT-2 XL · OpenWebText-trained avg perplexity | avg perplexity: 31.54 -> 25.93 | Improved |
| add two standard Transformer layers (parameter-matched) to GPT-2 XL (ablation) | GPT-2 XL · OpenWebText-trained avg perplexity | avg perplexity: 31.54 -> 31.57 | No clear effect |
| add two standard Transformer layers (parameter-matched) to GPT-2 Base (ablation) | GPT-2 Base · OpenWebText-trained avg perplexity | avg perplexity: 110.92 -> 107.77 | Improved |
| 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 -> 43.9 | Improved |
| add ContextLM to Pythia-70M | Pythia-70M · Pythia avg perplexity | avg perplexity: 297.72 -> 142.71 | Improved |
| add ContextLM to Pythia-1.4B | Pythia-1.4B · Pythia avg perplexity | avg perplexity: 9.74 -> 9.54 | Improved |
| add ContextLM to Pythia-70M | Pythia-70M · 9 downstream benchmarks avg accuracy (zero-shot) | avg accuracy: 34.6 -> 38.2 | Improved |
| add ContextLM to Pythia-1.4B | Pythia-1.4B · 9 downstream benchmarks avg accuracy (zero-shot) | avg accuracy: 54.1 -> 54.4 | Improved |
| add ContextLM to Pythia-70M, five-shot | Pythia-70M · 9 downstream benchmarks avg accuracy (five-shot) | avg accuracy: 32.6 -> 36.7 | Improved |
| fine-tune ContextLM-Pythia instead of Pythia on Alpaca | Pythia-1.4B · AlpacaEval 2.0 | win rate (%): 45.52 -> 54.48 | Improved |
| add ContextLM to Pythia-1.4B for long context | Pythia-1.4B · LongBench 0-4k | overall avg: 20.21 -> 23.32 | Improved |
Closest papers by meaning
- Towards Auto-Regressive Next-Token Prediction: In-context Learning Emerges from Generalization (ICLR 2025 · similarity 0.89)
- Better & Faster Large Language Models via Multi-token Prediction (ICML 2024 · similarity 0.89)
- Short-Context Dominance: How Much Local Context Natural Language Actually Needs? (NeurIPS 2026 · similarity 0.87)
- What is Wrong with Perplexity for Long-context Language Modeling? (ICLR 2025 · similarity 0.87)
- Beyond Next Token Prediction: Patch-Level Training for Large Language Models (ICLR 2025 · similarity 0.87)
- In-Context Language Learning: Architectures and Algorithms (ICML 2024 · similarity 0.87)
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.
