{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TEMDNXY57VF3LNA6NMTYGKCHDJ","short_pith_number":"pith:TEMDNXY5","schema_version":"1.0","canonical_sha256":"991836df1dfd4bb5b41e6b278328471a6595d4cf7d261b2f261837a7f31312e7","source":{"kind":"arxiv","id":"2506.20160","version":2},"attestation_state":"computed","paper":{"title":"AALC: Large Language Model Efficient Reasoning via Adaptive Accuracy-Length Control","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ali Payani, Ben Zhou, Quan Zhang, Ruochen Li, Ruosen Li, Xinya Du, Ziming Luo","submitted_at":"2025-06-25T06:29:18Z","abstract_excerpt":"Large reasoning models (LRMs) achieve impressive reasoning capabilities by generating lengthy chain-of-thoughts, but this \"overthinking\" incurs high latency and cost without commensurate accuracy gains. In this work, we introduce AALC, a lightweight, accuracy-aware length reward integrated into reinforcement learning that dynamically balances correctness and brevity during training. By incorporating validation accuracy into the reward and employing a smooth, dynamically scheduled length penalty, AALC delays length penalty until target performance is met. Through extensive experiments across st"},"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":"2506.20160","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-25T06:29:18Z","cross_cats_sorted":[],"title_canon_sha256":"f5d3b50f4a4c19615367623f79b1f33bca57e98b99c53d7579dd35f5adae46a2","abstract_canon_sha256":"ae0efe4d59daca0b34fd7e6555a5ec89086026d47232667eca66d882de5df5ff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:50:37.881490Z","signature_b64":"9Cw2ufrdEE3JJmuHJ1736JqhS9lDwd7qIX8Dj3T7mIlWDyxp3VMgcEfFuzaVbZaIvIBprFrT8KvB1lm3pvXwCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"991836df1dfd4bb5b41e6b278328471a6595d4cf7d261b2f261837a7f31312e7","last_reissued_at":"2026-07-05T11:50:37.880986Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:50:37.880986Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AALC: Large Language Model Efficient Reasoning via Adaptive Accuracy-Length Control","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ali Payani, Ben Zhou, Quan Zhang, Ruochen Li, Ruosen Li, Xinya Du, Ziming Luo","submitted_at":"2025-06-25T06:29:18Z","abstract_excerpt":"Large reasoning models (LRMs) achieve impressive reasoning capabilities by generating lengthy chain-of-thoughts, but this \"overthinking\" incurs high latency and cost without commensurate accuracy gains. In this work, we introduce AALC, a lightweight, accuracy-aware length reward integrated into reinforcement learning that dynamically balances correctness and brevity during training. By incorporating validation accuracy into the reward and employing a smooth, dynamically scheduled length penalty, AALC delays length penalty until target performance is met. Through extensive experiments across st"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.20160","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/2506.20160/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":"2506.20160","created_at":"2026-07-05T11:50:37.881052+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.20160v2","created_at":"2026-07-05T11:50:37.881052+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.20160","created_at":"2026-07-05T11:50:37.881052+00:00"},{"alias_kind":"pith_short_12","alias_value":"TEMDNXY57VF3","created_at":"2026-07-05T11:50:37.881052+00:00"},{"alias_kind":"pith_short_16","alias_value":"TEMDNXY57VF3LNA6","created_at":"2026-07-05T11:50:37.881052+00:00"},{"alias_kind":"pith_short_8","alias_value":"TEMDNXY5","created_at":"2026-07-05T11:50:37.881052+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":104,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22211","citing_title":"CLORE: Content-Level Optimization for Reasoning Efficiency","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ","json":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ.json","graph_json":"https://pith.science/api/pith-number/TEMDNXY57VF3LNA6NMTYGKCHDJ/graph.json","events_json":"https://pith.science/api/pith-number/TEMDNXY57VF3LNA6NMTYGKCHDJ/events.json","paper":"https://pith.science/paper/TEMDNXY5"},"agent_actions":{"view_html":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ","download_json":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ.json","view_paper":"https://pith.science/paper/TEMDNXY5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.20160&json=true","fetch_graph":"https://pith.science/api/pith-number/TEMDNXY57VF3LNA6NMTYGKCHDJ/graph.json","fetch_events":"https://pith.science/api/pith-number/TEMDNXY57VF3LNA6NMTYGKCHDJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ/action/storage_attestation","attest_author":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ/action/author_attestation","sign_citation":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ/action/citation_signature","submit_replication":"https://pith.science/pith/TEMDNXY57VF3LNA6NMTYGKCHDJ/action/replication_record"}},"created_at":"2026-07-05T11:50:37.881052+00:00","updated_at":"2026-07-05T11:50:37.881052+00:00"}