POLARIS: Guiding Small Models to Write Long Stories
Quick Take
POLARIS is a new training approach for small models like Qwen3.5-9B, enhancing long-form story generation by incorporating a frontier LLM judge and human-written story references. It outperforms base models and competes with larger models, maintaining quality even for prompts up to three times the training length.
Key Points
- POLARIS uses a frontier LLM judge for structured story quality assessment.
- Training involved 1.4K prompt-story pairs and 4 A100 GPUs.
- POLARIS-9B closely follows length instructions, outperforming Qwen3.5-9B.
- It maintains quality on prompts requesting stories up to 12k words.
- Length generalization is a key metric for evaluating creative-writing models.
Article Content
From source RSS / original summaryarXiv:2606. 04095v1 Announce Type: new Abstract: 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.
Reader Mode unavailable (could not extract clean content).
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Time to REFLECT: Can We Trust LLM Judges for Evidence-based Research Agents?
The REFLECT benchmark reveals that current LLM judges are unreliable, achieving below 55% accuracy in evaluating reasoning and evidence use, highlighting the need for improved evaluation methods for deep research agents.