{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EGE5PMLFDSDP4V7DFA2JKA4YDS","short_pith_number":"pith:EGE5PMLF","schema_version":"1.0","canonical_sha256":"2189d7b1651c86fe57e328349503981c961211c2709fe4f2bb4a4e663864b06e","source":{"kind":"arxiv","id":"2512.21002","version":3},"attestation_state":"computed","paper":{"title":"Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ata Fatahibaarzi, Hejian Sang, Muhammad Abdul-Mageed, Qingquan Song, Shao Tang, Vignesh Kothapalli, Wei-Rui Chen, Zhipeng Wang","submitted_at":"2025-12-24T06:57:35Z","abstract_excerpt":"Distilling the capabilities from a large reasoning model (LRM) to a smaller student model often involves training on substantial amounts of reasoning data. However, knowledge distillation (KD) over lengthy sequences with prompt (P), chain-of-thought (CoT), and answer (A) sections makes the process computationally expensive. In this work, we investigate how the allocation of supervision across different sections (P, CoT, A) affects student performance. Our analysis shows that selective KD over only the CoT tokens can be effective when the prompt and answer information is encompassed by it. Buil"},"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":"2512.21002","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-12-24T06:57:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"410c6478f72f6f5368c6e945bef12bc5dae26d4ca46198adb0e79a582abef08c","abstract_canon_sha256":"2086d7d5220c00c5296519202e8aca56573f70f628f98993ebf98805c5d9648c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-01T01:17:44.232042Z","signature_b64":"i/3KKS0zzs4Unx534AE136l7LzkATXTfRC+wu+FUIh1C/vn7ljVQwAEzZObzYUZSbcOUyVf1iEphdoy8LLz4CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2189d7b1651c86fe57e328349503981c961211c2709fe4f2bb4a4e663864b06e","last_reissued_at":"2026-07-01T01:17:44.231064Z","signature_status":"signed_v1","first_computed_at":"2026-07-01T01:17:44.231064Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ata Fatahibaarzi, Hejian Sang, Muhammad Abdul-Mageed, Qingquan Song, Shao Tang, Vignesh Kothapalli, Wei-Rui Chen, Zhipeng Wang","submitted_at":"2025-12-24T06:57:35Z","abstract_excerpt":"Distilling the capabilities from a large reasoning model (LRM) to a smaller student model often involves training on substantial amounts of reasoning data. However, knowledge distillation (KD) over lengthy sequences with prompt (P), chain-of-thought (CoT), and answer (A) sections makes the process computationally expensive. In this work, we investigate how the allocation of supervision across different sections (P, CoT, A) affects student performance. Our analysis shows that selective KD over only the CoT tokens can be effective when the prompt and answer information is encompassed by it. Buil"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.21002","kind":"arxiv","version":3},"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/2512.21002/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":"2512.21002","created_at":"2026-07-01T01:17:44.231128+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.21002v3","created_at":"2026-07-01T01:17:44.231128+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.21002","created_at":"2026-07-01T01:17:44.231128+00:00"},{"alias_kind":"pith_short_12","alias_value":"EGE5PMLFDSDP","created_at":"2026-07-01T01:17:44.231128+00:00"},{"alias_kind":"pith_short_16","alias_value":"EGE5PMLFDSDP4V7D","created_at":"2026-07-01T01:17:44.231128+00:00"},{"alias_kind":"pith_short_8","alias_value":"EGE5PMLF","created_at":"2026-07-01T01:17:44.231128+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":69,"is_internal_anchor":true},{"citing_arxiv_id":"2603.05433","citing_title":"CRISP: Compressed Reasoning via Iterative Self-Policy Distillation","ref_index":2,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS","json":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS.json","graph_json":"https://pith.science/api/pith-number/EGE5PMLFDSDP4V7DFA2JKA4YDS/graph.json","events_json":"https://pith.science/api/pith-number/EGE5PMLFDSDP4V7DFA2JKA4YDS/events.json","paper":"https://pith.science/paper/EGE5PMLF"},"agent_actions":{"view_html":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS","download_json":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS.json","view_paper":"https://pith.science/paper/EGE5PMLF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.21002&json=true","fetch_graph":"https://pith.science/api/pith-number/EGE5PMLFDSDP4V7DFA2JKA4YDS/graph.json","fetch_events":"https://pith.science/api/pith-number/EGE5PMLFDSDP4V7DFA2JKA4YDS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS/action/storage_attestation","attest_author":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS/action/author_attestation","sign_citation":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS/action/citation_signature","submit_replication":"https://pith.science/pith/EGE5PMLFDSDP4V7DFA2JKA4YDS/action/replication_record"}},"created_at":"2026-07-01T01:17:44.231128+00:00","updated_at":"2026-07-01T01:17:44.231128+00:00"}