{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6DIN4UJKWTGBETNDOFDMBCXCV5","short_pith_number":"pith:6DIN4UJK","schema_version":"1.0","canonical_sha256":"f0d0de512ab4cc124da37146c08ae2af5cc75c2fd4d923b0789f2303bb9c510b","source":{"kind":"arxiv","id":"2506.09991","version":2},"attestation_state":"computed","paper":{"title":"Multiverse: Your Language Models Secretly Decide How to Parallelize and Merge Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Beidi Chen, Hongyi Liu, Tianqi Chen, Xinyu Yang, Yuwei An","submitted_at":"2025-06-11T17:59:23Z","abstract_excerpt":"Autoregressive Large Language Models (AR-LLMs) frequently exhibit implicit parallelism in sequential generation. Inspired by this, we introduce Multiverse, a new generative model that enables natively parallel generation. Multiverse internalizes a MapReduce paradigm, generating automatically through three stages: (i) a Map stage for adaptive task decomposition, (ii) a Process stage for parallel subtask execution, and (iii) a Reduce stage for lossless result synthesis. Next, we build a real-world Multiverse reasoning model with co-design of data, algorithm, and system, enabling rapid and seamle"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2506.09991","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-11T17:59:23Z","cross_cats_sorted":[],"title_canon_sha256":"d762fec25be2041c2cd1f3a6192a6db5ecf3da5b45152e8aa0fd2759c7f1fc3d","abstract_canon_sha256":"1c8c0f92d8ceff11d82f5397ef36d952dbd410878cc72a1576efeaa769d9f2dd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:56.683010Z","signature_b64":"BZQ2DKBNfiWDzpnWoDXGQD7WJvRwfsRIRvFCV3AkXwDeTD9DK1km0biBhcgSRfo2mSTfK751+CUreBAPMhGcDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f0d0de512ab4cc124da37146c08ae2af5cc75c2fd4d923b0789f2303bb9c510b","last_reissued_at":"2026-07-05T11:20:56.682530Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:56.682530Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multiverse: Your Language Models Secretly Decide How to Parallelize and Merge Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Beidi Chen, Hongyi Liu, Tianqi Chen, Xinyu Yang, Yuwei An","submitted_at":"2025-06-11T17:59:23Z","abstract_excerpt":"Autoregressive Large Language Models (AR-LLMs) frequently exhibit implicit parallelism in sequential generation. Inspired by this, we introduce Multiverse, a new generative model that enables natively parallel generation. Multiverse internalizes a MapReduce paradigm, generating automatically through three stages: (i) a Map stage for adaptive task decomposition, (ii) a Process stage for parallel subtask execution, and (iii) a Reduce stage for lossless result synthesis. Next, we build a real-world Multiverse reasoning model with co-design of data, algorithm, and system, enabling rapid and seamle"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09991","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2506.09991/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2506.09991","created_at":"2026-07-05T11:20:56.682606+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.09991v2","created_at":"2026-07-05T11:20:56.682606+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09991","created_at":"2026-07-05T11:20:56.682606+00:00"},{"alias_kind":"pith_short_12","alias_value":"6DIN4UJKWTGB","created_at":"2026-07-05T11:20:56.682606+00:00"},{"alias_kind":"pith_short_16","alias_value":"6DIN4UJKWTGBETND","created_at":"2026-07-05T11:20:56.682606+00:00"},{"alias_kind":"pith_short_8","alias_value":"6DIN4UJK","created_at":"2026-07-05T11:20:56.682606+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03102","citing_title":"Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling","ref_index":102,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15177","citing_title":"OpenDeepThink: Parallel Reasoning via Bradley-Terry Aggregation","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2512.07461","citing_title":"Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15177","citing_title":"OpenDeepThink: Parallel Reasoning via Bradley-Terry Aggregation","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06914","citing_title":"Regulating Branch Parallelism in LLM Serving","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5","json":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5.json","graph_json":"https://pith.science/api/pith-number/6DIN4UJKWTGBETNDOFDMBCXCV5/graph.json","events_json":"https://pith.science/api/pith-number/6DIN4UJKWTGBETNDOFDMBCXCV5/events.json","paper":"https://pith.science/paper/6DIN4UJK"},"agent_actions":{"view_html":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5","download_json":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5.json","view_paper":"https://pith.science/paper/6DIN4UJK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.09991&json=true","fetch_graph":"https://pith.science/api/pith-number/6DIN4UJKWTGBETNDOFDMBCXCV5/graph.json","fetch_events":"https://pith.science/api/pith-number/6DIN4UJKWTGBETNDOFDMBCXCV5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5/action/storage_attestation","attest_author":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5/action/author_attestation","sign_citation":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5/action/citation_signature","submit_replication":"https://pith.science/pith/6DIN4UJKWTGBETNDOFDMBCXCV5/action/replication_record"}},"created_at":"2026-07-05T11:20:56.682606+00:00","updated_at":"2026-07-05T11:20:56.682606+00:00"}