← 研踪 PaperTrailarXiv:2611.25003 · v1
A Model of Diverse Sampling from Language Models
首次提交 2026-11-19 01:56
版本更新 2026-11-19 01:56
雷达观测 2026-11-21 12:18
摘要
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- 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 -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 -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@.
作者备注:NeurIPS 2026 paper
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-19 01:56 · 最新观测版本
A Model of Diverse Sampling from Language Models
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- 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 -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 -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@.
查看此版本解读 →查看此版本原文 ↗