Scaling Limits of Long-Context Transformers
cs.LG, cs.AI, math.PR, math.ST, stat.TH
Submitted: 2026-05-08
Updated: 2026-11-03
Comments: 41 pages, 4 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Qwen Technical Report
- Longformer: The Long-Document Transformer
- Overcoming a Theoretical Limitation of Self-Attention
- Quantitative Clustering in Mean-Field Transformer Models
- Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context
- Scalable-Softmax Is Superior for Attention
- TabICLv2: A better, faster, scalable, and open tabular foundation model
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis