# POLARIS: Guiding Small Models to Write Long Stories

arXiv: https://arxiv.org/abs/2606.04095 · HTML: https://mlsanity.com/paper/2606.04095/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 50 references that are in the map.
- First posted on arXiv 2 Jun 2026.

## Builds on (most cited first, 27 of 27)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [A Survey on LLM-as-a-Judge](https://mlsanity.com/paper/2411.15594/) (2024)
- [Large Language Models are not Fair Evaluators](https://mlsanity.com/paper/2305.17926/) (2023)
- [PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel](https://mlsanity.com/paper/2304.11277/) (2023)
- [Prometheus: Inducing Fine-grained Evaluation Capability in Language Models](https://mlsanity.com/paper/2310.08491/) (2023)
- [LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods](https://mlsanity.com/paper/2412.05579/) (2024)
- [Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Models](https://mlsanity.com/paper/2405.01535/) (2024)
- [Learning to Reason under Off-Policy Guidance](https://mlsanity.com/paper/2504.14945/) (2025)
- [LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs](https://mlsanity.com/paper/2408.07055/) (2024)
- [BREAD: Branched Rollouts from Expert Anchors Bridge SFT & RL for Reasoning](https://mlsanity.com/paper/2506.17211/) (2025)
- [Diverse Preference Optimization](https://mlsanity.com/paper/2501.18101/) (2025)
- [Writing-Zero: Bridge the Gap Between Non-verifiable Tasks and Verifiable Rewards](https://mlsanity.com/paper/2506.00103/) (2025)
- [LitBench: A Benchmark and Dataset for Reliable Evaluation of Creative Writing](https://mlsanity.com/paper/2507.00769/) (2025)
- [Measuring memorization in language models via probabilistic extraction](https://mlsanity.com/paper/2410.19482/) (2024)
- [Learning to Reason for Long-Form Story Generation](https://mlsanity.com/paper/2503.22828/) (2025)
- [Creative Preference Optimization](https://mlsanity.com/paper/2505.14442/) (2025)
- [RLMR: Reinforcement Learning with Mixed Rewards for Creative Writing](https://mlsanity.com/paper/2508.18642/) (2025)
- [Igniting Creative Writing in Small Language Models: LLM-as-a-Judge versus Multi-Agent Refined Rewards](https://mlsanity.com/paper/2508.21476/) (2025)
- [Writing-RL: Advancing Long-form Writing via Adaptive Curriculum Reinforcement Learning](https://mlsanity.com/paper/2506.05760/) (2025)
- [StoryScope: Investigating idiosyncrasies in AI fiction](https://mlsanity.com/paper/2604.03136/) (2026)
- [SWAG: Storytelling With Action Guidance](https://mlsanity.com/paper/2402.03483/) (2024)
- [DPWriter: Reinforcement Learning with Diverse Planning Branching for Creative Writing](https://mlsanity.com/paper/2601.09609/) (2026)
- [R2-Write: Reflection and Revision for Open-Ended Writing with Deep Reasoning](https://mlsanity.com/paper/2604.03004/) (2026)
- [Rewarding Creativity: A Human-Aligned Generative Reward Model for Reinforcement Learning in Storytelling](https://mlsanity.com/paper/2601.07149/) (2026)
- [Writer-R1: Enhancing Generative Writing in LLMs via Memory-augmented Replay Policy Optimization](https://mlsanity.com/paper/2603.15061/) (2026)

## Abstract (arXiv)

Small open-weight models struggle at long-form creative writing: their generated stories either fall far short of the requested length, or their quality significantly degrades as length increases, especially when compared to frontier models. We present POLARIS (Policy Optimization with LLM-as-a-judge rewards and Anchored-Reference Injection for Storywriting), a lower-compute GRPO recipe with two key ingredients: a frontier LLM judge with a structured Story Quality rubric as the online reward, and human-reference injection (HRI), where a teacher-forced human-written story serves as a high-reward anchor within each GRPO group. By applying our training recipe to Qwen3.5-9B, using a dataset of approximately 1.4K prompt-story pairs derived from 100 short-story anthologies and 4 A100 GPUs, we obtain POLARIS-9B. Across five benchmarks spanning in-distribution and out-of-distribution prompts and rubrics, POLARIS-9B is competitive with much larger open-weight models while following length instructions more closely. A blinded human evaluation confirms that POLARIS-9B is preferred to the base Qwen3.5-9B and on par with Qwen3.5-27B. Despite training only on stories up to 4k words, POLARIS-9B preserves quality on prompts requesting stories up to 3 times the training length, a regime where most open-weight models degrade substantially in quality, length adherence, or both. More broadly, our results suggest that length generalization is a meaningful stress test for creative-writing models and a useful lens for distinguishing otherwise close models.