{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MPW4LJNO5YQ65OOANWALZVZUEE","short_pith_number":"pith:MPW4LJNO","schema_version":"1.0","canonical_sha256":"63edc5a5aeee21eeb9c06d80bcd734213398385a61f00b497285b816ebec88d6","source":{"kind":"arxiv","id":"2608.04962","version":1},"attestation_state":"computed","paper":{"title":"SpecRoll: Fast-Slow Verifier-Feedback Adaptation for Speculative Reinforcement Learning Rollouts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Duy Tung Doan, Khac-Hoai Nam Bui, Nhat Minh Pham, Thi Duyen Ngo, Vinh Van Nguyen","submitted_at":"2026-08-05T15:32:27Z","abstract_excerpt":"Reinforcement learning (RL) post-training improves the reasoning capabilities of large language models, but autoregressive rollout generation remains a major efficiency bottleneck. Speculative decoding can accelerate generation, yet applying it during RL is difficult because the target policy continually evolves: static proposers become stale, while frequent drafter updates add substantial overhead. We introduce SpecRoll, a speculative rollout engine that preserves the target model's sampling distribution while adapting at two timescales. Lightweight future-token heads generate parallel propos"},"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":"2608.04962","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-08-05T15:32:27Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"1b318fa44055ba1212a3140c519a6d65a32ee49a6055fd1255efbf58298361c2","abstract_canon_sha256":"6ab90c35fed54dc212c56d28b165651da1fd5da484aa6ee18a4ae0c805b4d977"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-06T01:47:55.110097Z","signature_b64":"INab1mouje/iFDiUolISHqfa4WSun8yYs+G1XCplqLXZs7DmVAkPO7dzYdujT2/upS0Dd+h9l5SmcPkToNU3Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"63edc5a5aeee21eeb9c06d80bcd734213398385a61f00b497285b816ebec88d6","last_reissued_at":"2026-08-06T01:47:55.108700Z","signature_status":"signed_v1","first_computed_at":"2026-08-06T01:47:55.108700Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SpecRoll: Fast-Slow Verifier-Feedback Adaptation for Speculative Reinforcement Learning Rollouts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Duy Tung Doan, Khac-Hoai Nam Bui, Nhat Minh Pham, Thi Duyen Ngo, Vinh Van Nguyen","submitted_at":"2026-08-05T15:32:27Z","abstract_excerpt":"Reinforcement learning (RL) post-training improves the reasoning capabilities of large language models, but autoregressive rollout generation remains a major efficiency bottleneck. Speculative decoding can accelerate generation, yet applying it during RL is difficult because the target policy continually evolves: static proposers become stale, while frequent drafter updates add substantial overhead. We introduce SpecRoll, a speculative rollout engine that preserves the target model's sampling distribution while adapting at two timescales. Lightweight future-token heads generate parallel propos"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.04962","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/2608.04962/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":"2608.04962","created_at":"2026-08-06T01:47:55.110508+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.04962v1","created_at":"2026-08-06T01:47:55.110508+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.04962","created_at":"2026-08-06T01:47:55.110508+00:00"},{"alias_kind":"pith_short_12","alias_value":"MPW4LJNO5YQ6","created_at":"2026-08-06T01:47:55.110508+00:00"},{"alias_kind":"pith_short_16","alias_value":"MPW4LJNO5YQ65OOA","created_at":"2026-08-06T01:47:55.110508+00:00"},{"alias_kind":"pith_short_8","alias_value":"MPW4LJNO","created_at":"2026-08-06T01:47:55.110508+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/MPW4LJNO5YQ65OOANWALZVZUEE","json":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE.json","graph_json":"https://pith.science/api/pith-number/MPW4LJNO5YQ65OOANWALZVZUEE/graph.json","events_json":"https://pith.science/api/pith-number/MPW4LJNO5YQ65OOANWALZVZUEE/events.json","paper":"https://pith.science/paper/MPW4LJNO"},"agent_actions":{"view_html":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE","download_json":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE.json","view_paper":"https://pith.science/paper/MPW4LJNO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.04962&json=true","fetch_graph":"https://pith.science/api/pith-number/MPW4LJNO5YQ65OOANWALZVZUEE/graph.json","fetch_events":"https://pith.science/api/pith-number/MPW4LJNO5YQ65OOANWALZVZUEE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE/action/storage_attestation","attest_author":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE/action/author_attestation","sign_citation":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE/action/citation_signature","submit_replication":"https://pith.science/pith/MPW4LJNO5YQ65OOANWALZVZUEE/action/replication_record"}},"created_at":"2026-08-06T01:47:55.110508+00:00","updated_at":"2026-08-06T01:47:55.110508+00:00"}