{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HNTXAGV5VUEERYS3UFPQ2G4P5L","short_pith_number":"pith:HNTXAGV5","schema_version":"1.0","canonical_sha256":"3b67701abdad0848e25ba15f0d1b8feaf79dd2bdb8517261e07f18558279319a","source":{"kind":"arxiv","id":"2606.08446","version":1},"attestation_state":"computed","paper":{"title":"Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aram Galstyan, Beidi Chen, Haizhong Zheng, Ranajoy Sadhukhan, Sai Muralidhar Jayanthi, Saket Dingliwal, Souvik Kundu, Yang Zhou, Zhaofeng Sun, Zhuoming Chen","submitted_at":"2026-06-07T04:24:45Z","abstract_excerpt":"Despite being powerful, reinforcement learning with verifiable rewards (RLVR) induces extremely long COT, making it computationally expensive. Since RLVR per-step cost is dominated by long-context rollout generation, sparse attention offers a promising way to accelerate dense rollout. However, sparse rollouts require a delicate stability-efficiency tradeoff: overly aggressive sparsity causes collapse, while overly lenient sparsity gives insufficient speedup. In this work, we study this tradeoff through sparse-to-dense actor-policy mismatch. We first observe that sparse rollout collapse is not "},"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":"2606.08446","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-07T04:24:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2b302eb64bb662dc8245267ec91c6138a9bcb45bcc853def10ca9cd108a0b5f3","abstract_canon_sha256":"bf5ef923f92ada9dbed10caecddba30504cef770c5096c1a3200533ba5a05c2f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:05:36.867528Z","signature_b64":"Fk+3I0BkVT58hNCD06FWeD+0H0Ku8T5bK9atUtWXL//T7TlcKKxFl1wUhPRbE4qmfs6v1J17PN7SjY5mERYhAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3b67701abdad0848e25ba15f0d1b8feaf79dd2bdb8517261e07f18558279319a","last_reissued_at":"2026-06-09T01:05:36.867141Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:05:36.867141Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aram Galstyan, Beidi Chen, Haizhong Zheng, Ranajoy Sadhukhan, Sai Muralidhar Jayanthi, Saket Dingliwal, Souvik Kundu, Yang Zhou, Zhaofeng Sun, Zhuoming Chen","submitted_at":"2026-06-07T04:24:45Z","abstract_excerpt":"Despite being powerful, reinforcement learning with verifiable rewards (RLVR) induces extremely long COT, making it computationally expensive. Since RLVR per-step cost is dominated by long-context rollout generation, sparse attention offers a promising way to accelerate dense rollout. However, sparse rollouts require a delicate stability-efficiency tradeoff: overly aggressive sparsity causes collapse, while overly lenient sparsity gives insufficient speedup. In this work, we study this tradeoff through sparse-to-dense actor-policy mismatch. We first observe that sparse rollout collapse is not "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.08446","kind":"arxiv","version":1},"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/2606.08446/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":"2606.08446","created_at":"2026-06-09T01:05:36.867203+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.08446v1","created_at":"2026-06-09T01:05:36.867203+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.08446","created_at":"2026-06-09T01:05:36.867203+00:00"},{"alias_kind":"pith_short_12","alias_value":"HNTXAGV5VUEE","created_at":"2026-06-09T01:05:36.867203+00:00"},{"alias_kind":"pith_short_16","alias_value":"HNTXAGV5VUEERYS3","created_at":"2026-06-09T01:05:36.867203+00:00"},{"alias_kind":"pith_short_8","alias_value":"HNTXAGV5","created_at":"2026-06-09T01:05:36.867203+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L","json":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L.json","graph_json":"https://pith.science/api/pith-number/HNTXAGV5VUEERYS3UFPQ2G4P5L/graph.json","events_json":"https://pith.science/api/pith-number/HNTXAGV5VUEERYS3UFPQ2G4P5L/events.json","paper":"https://pith.science/paper/HNTXAGV5"},"agent_actions":{"view_html":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L","download_json":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L.json","view_paper":"https://pith.science/paper/HNTXAGV5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.08446&json=true","fetch_graph":"https://pith.science/api/pith-number/HNTXAGV5VUEERYS3UFPQ2G4P5L/graph.json","fetch_events":"https://pith.science/api/pith-number/HNTXAGV5VUEERYS3UFPQ2G4P5L/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L/action/storage_attestation","attest_author":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L/action/author_attestation","sign_citation":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L/action/citation_signature","submit_replication":"https://pith.science/pith/HNTXAGV5VUEERYS3UFPQ2G4P5L/action/replication_record"}},"created_at":"2026-06-09T01:05:36.867203+00:00","updated_at":"2026-06-09T01:05:36.867203+00:00"}