# PolicyLong: Towards On-Policy Context Extension

arXiv: https://arxiv.org/abs/2604.07809 · HTML: https://mlsanity.com/paper/2604.07809/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 29 references that are in the map.
- First posted on arXiv 9 Apr 2026.

## Builds on (most cited first, 24 of 24)

- [A Simple Framework for Contrastive Learning of Visual Representations](https://mlsanity.com/paper/2002.05709/) (2020)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Dense Passage Retrieval for Open-Domain Question Answering](https://mlsanity.com/paper/2004.04906/) (2020)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Billion-scale similarity search with GPUs](https://mlsanity.com/paper/1702.08734/) (2017)
- [RULER: What's the Real Context Size of Your Long-Context Language Models?](https://mlsanity.com/paper/2404.06654/) (2024)
- [Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval](https://mlsanity.com/paper/2007.00808/) (2020)
- [Contrastive Learning with Hard Negative Samples](https://mlsanity.com/paper/2010.04592/) (2020)
- [RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering](https://mlsanity.com/paper/2010.08191/) (2020)
- [LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks](https://mlsanity.com/paper/2412.15204/) (2024)
- [Reinforcement Learning via Self-Distillation](https://mlsanity.com/paper/2601.20802/) (2026)
- [Effective Long-Context Scaling of Foundation Models](https://mlsanity.com/paper/2309.16039/) (2023)
- [Self-Distillation Enables Continual Learning](https://mlsanity.com/paper/2601.19897/) (2026)
- [jina-embeddings-v3: Multilingual Embeddings With Task LoRA](https://mlsanity.com/paper/2409.10173/) (2024)
- [HELMET: How to Evaluate Long-Context Language Models Effectively and Thoroughly](https://mlsanity.com/paper/2410.02694/) (2024)
- [In-context Pretraining: Language Modeling Beyond Document Boundaries](https://mlsanity.com/paper/2310.10638/) (2023)
- [What is Wrong with Perplexity for Long-context Language Modeling?](https://mlsanity.com/paper/2410.23771/) (2024)
- [POPE: Learning to Reason on Hard Problems via Privileged On-Policy Exploration](https://mlsanity.com/paper/2601.18779/) (2026)
- [NExtLong: Toward Effective Long-Context Training without Long Documents](https://mlsanity.com/paper/2501.12766/) (2025)
- [Quest: Query-centric Data Synthesis Approach for Long-context Scaling of Large Language Model](https://mlsanity.com/paper/2405.19846/) (2024)
- [LongMagpie: A Self-synthesis Method for Generating Large-scale Long-context Instructions](https://mlsanity.com/paper/2505.17134/) (2025)
- [EntropyLong: Effective Long-Context Training via Predictive Uncertainty](https://mlsanity.com/paper/2510.02330/) (2025)
- [LiteLong: Resource-Efficient Long-Context Data Synthesis for LLMs](https://mlsanity.com/paper/2509.15568/) (2025)

## Abstract (arXiv)

Extending LLM context windows is hindered by scarce high-quality long-context data. Recent methods synthesize data with genuine long-range dependencies via information-theoretic verification, selecting contexts that reduce a base model's predictive entropy. However, their single-pass offline construction with a fixed model creates a fundamental off-policy gap: the static screening landscape misaligns with the model's evolving capabilities, causing the training distribution to drift. We propose PolicyLong, shifting data construction towards a dynamic on-policy paradigm. By iteratively re-executing data screening (entropy computation, retrieval, and verification) using the current model, PolicyLong ensures the training distribution tracks evolving capabilities, yielding an emergent self-curriculum. Crucially, both positive and hard negative contexts derive from the current model's entropy landscape, co-evolving what the model learns to exploit and resist. Experiments on RULER, HELMET, and LongBench-v2 (Qwen2.5-3B) show PolicyLong consistently outperforms EntropyLong and NExtLong, with gains growing at longer contexts (e.g., +2.54 at 128K on RULER), confirming the value of on-policy data evolution.