AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Flow of Spans: Generalizing Language Models to Dynamic Span-Vocabulary via GFlowNets

Bo Xue, Yunchong Song, Fanghao Shao, Xuekai Zhu, Lin Chen, Luoyi Fu, Xinbing Wang, Zhouhan Lin

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Controlled text generation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
model the state space as a DAG with phrases instead of a tree (ablation)FoSS · WikiText-103 in-domain, nucleusMAUVE: 29.61 -> 31.65Improved
model the state space as a DAG with phrases instead of a tree (ablation)FoSS · WikiText-103 in-domain, nucleusDiversity: 65.72 -> 92.48Improved
add preference model reward (ablation)FoSS · WikiText-103 in-domain, nucleusMAUVE: 28.25 -> 31.65Improved
add language model reward (ablation)FoSS · WikiText-103 in-domain, nucleusMAUVE: 29.09 -> 31.65Improved
add language model reward on top of preference model (ablation)FoSS · WikiText-103 in-domain, nucleusDiversity: 92.77 -> 92.48No clear effect
add language model reward (ablation)FoSS · Law-MT out-of-domain, nucleusMAUVE: 29.79 -> 32.17Improved
use DAG span state space with GFlowNets instead of tree GFlowNets-FTFoSS · WikiText-103 in-domain, nucleusMAUVE: 29.61 -> 31.65Improved
use FoSS instead of CoGFoSS · WikiText-103 in-domain, nucleusMAUVE: 26.14 -> 31.65Improved
use FoSS instead of fine-tuned TransformerFoSS · Law-MT out-of-domain, nucleusMAUVE: 26.85 -> 32.17Improved
use larger En-Wiki data store without retrainingFoSS · WikiText-103 test, nucleusMAUVE: 31.65 -> 33.79Improved
use FoSS instead of fine-tuned TransformerFoSS · OpenBookQAaccuracy (%): 22.71 -> 26.20Improved
use FoSS instead of CoGFoSS · MedMCQAaccuracy (%): 27.55 -> 27.44Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy