{"slug": "a-model-of-diverse-sampling-from-language-models", "title": "A Model of Diverse Sampling from Language Models", "summary": "A new arXiv paper (2611.25003v1) proposes PoolDPP, a method for drawing diverse sets of outputs from language models by sampling a size-K subset of complete strings from a determinantal point process, where likelihood encodes item quality and embedding geometry encodes repulsion between similar outputs. Because exact inference over all strings is intractable, PoolDPP draws a finite candidate pool from a tractable proposal, importance-weights candidates toward the target quality, and runs a K-DPP sampler on the induced pool kernel, with a greedy MAP variant for deterministic selection; the authors prove the KL divergence between the sampler's output and the target K-DPP decays as O(1/N) in pool size. Empirically, PoolDPP extends the quality-diversity frontier on open-ended generation and the precision-recall frontier on ambiguous text-to-SQL (Ambrosia), and on mathematical reasoning (GSM8K) it raises diversity at about the same pass@K.", "body_md": "arXiv:2611.25003v1 Announce Type: cross \nAbstract: Language models often produce high-quality individual samples but poor sets, with repeated samples clustering around the same semantic modes. We formalize diverse generation as sampling a size-$K$ subset of complete strings from a determinantal point process (DPP), where the likelihood encodes item quality, and the embedding geometry encodes repulsion between similar outputs. Exact inference over all strings is intractable, so we propose PoolDPP: draw a finite candidate pool from a tractable proposal, importance-weight the candidates toward the target quality, and run a $K$-DPP sampler on the induced pool kernel; a greedy MAP variant provides deterministic selection. We prove a finite-sample bound with an explicit constant: the KL divergence between the sampler's output and the target $K$-DPP decays as O(1/N) in the pool size. Empirically, PoolDPP extends the quality--diversity frontier on open-ended generation and the precision--recall frontier on ambiguous text-to-SQL (Ambrosia), and on mathematical reasoning (GSM8K), it raises diversity at about the same pass@$K$.", "url": "https://wpnews.pro/news/a-model-of-diverse-sampling-from-language-models", "canonical_source": "https://arxiv.org/abs/2611.25003", "published_at": "2026-11-20 05:00:00+00:00", "updated_at": "2026-11-20 05:12:47.334778+00:00", "lang": "en", "topics": ["large-language-models", "machine-learning", "ai-research", "natural-language-processing"], "entities": ["PoolDPP", "Ambrosia", "GSM8K", "arXiv"], "also_reported_by": [], "takeaway": "PoolDPP's KL divergence from the target K-DPP decays as O(1/N) in candidate pool size, and it raises diversity at roughly the same pass@K on GSM8K while extending the quality-diversity frontier on open-ended generation.", "recommendation": {"advice": "When generating K outputs, sample a finite candidate pool, importance-weight toward quality, and run a K-DPP sampler on the pool kernel to trade off quality and diversity; the source bounds divergence at O(1/N) in pool size.", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/a-model-of-diverse-sampling-from-language-models", "markdown": "https://wpnews.pro/news/a-model-of-diverse-sampling-from-language-models.md", "text": "https://wpnews.pro/news/a-model-of-diverse-sampling-from-language-models.txt", "jsonld": "https://wpnews.pro/news/a-model-of-diverse-sampling-from-language-models.jsonld"}}