{"paper":{"title":"Language Modeling with Hyperspherical Flows","license":"http://creativecommons.org/licenses/by/4.0/","headline":"S-FLM rotates vectors on the hypersphere to let continuous flow language models approach masked diffusion on reasoning tasks.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Caglar Gulcehre, Justin Deschenaux","submitted_at":"2026-05-11T18:32:32Z","abstract_excerpt":"Discrete Diffusion Language Models progressed rapidly as an alternative to autoregressive (AR) models, motivated by their parallel generation abilities. However, for tractability, discrete diffusion models sample from a factorized distribution, which is less expressive than AR. Recent Flow Language Models (FLMs) apply continuous flows to language, transporting noise to data with a deterministic ODE that avoids factorized sampling. FLMs operate on one-hot vectors whose dimension scales with the vocabulary size, making FLMs costly to train. Moreover, since all distinct one-hot embeddings are equ"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"S-FLM substantially improves continuous flow language models on large-vocabulary reasoning and closes the gap to masked diffusion under standard-temperature sampling (T=1), while a gap remains under optimized low-temperature (T=0.1) decoding.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That rotating vectors on the hypersphere along a learned velocity field provides a semantically meaningful transport from noise to data, unlike the equidistant one-hot vectors used in prior FLMs.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"S-FLM rotates vectors on a hypersphere using a learned velocity field to generate language sequences, improving continuous flow models on large-vocabulary reasoning and closing the gap to masked diffusion at standard sampling temperature.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"S-FLM rotates vectors on the hypersphere to let continuous flow language models approach masked diffusion on reasoning tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d7eaa57b31220fabfe04bc9dc0d78c8c9d23fc50fd05547c8eb9168c9a0b655e"},"source":{"id":"2605.11125","kind":"arxiv","version":2},"verdict":{"id":"7bdfe128-8a58-47a8-a7d7-28cdece271b0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T06:49:29.117985Z","strongest_claim":"S-FLM substantially improves continuous flow language models on large-vocabulary reasoning and closes the gap to masked diffusion under standard-temperature sampling (T=1), while a gap remains under optimized low-temperature (T=0.1) decoding.","one_line_summary":"S-FLM rotates vectors on a hypersphere using a learned velocity field to generate language sequences, improving continuous flow models on large-vocabulary reasoning and closing the gap to masked diffusion at standard sampling temperature.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That rotating vectors on the hypersphere along a learned velocity field provides a semantically meaningful transport from noise to data, unlike the equidistant one-hot vectors used in prior FLMs.","pith_extraction_headline":"S-FLM rotates vectors on the hypersphere to let continuous flow language models approach masked diffusion on reasoning tasks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11125/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:34:57.997291Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:16.785500Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:46:22.317458Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"65f1df7664bc36aef18680f71de21c53c557a193e6f0b0bbb155a68e571512a0"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2d14b24349c1215935cae1ca7c3c9d2be462079915fed421e3bfcb02d3ebc7b2"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}