{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:OKLSLVT4VHR5HMYKJK3QZXTYV6","short_pith_number":"pith:OKLSLVT4","schema_version":"1.0","canonical_sha256":"729725d67ca9e3d3b30a4ab70cde78afabe1291136e70c0098279d8872e1f2c7","source":{"kind":"arxiv","id":"2602.02244","version":3},"attestation_state":"computed","paper":{"title":"Entropy-Preserving Supervised Fine-Tuning via Adaptive Self-Distillation for Large Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Dapeng Wu, Hao Gu, Hao Wang, Hongming Piao, Kaixiong Gong, Sirui Han, Xiangyu Yue, Yike Guo, Yuxiao Ye","submitted_at":"2026-02-02T15:53:55Z","abstract_excerpt":"The standard post-training recipe for large reasoning models, supervised fine-tuning followed by reinforcement learning (SFT-then-RL), may limit the benefits of the RL stage: while SFT imitates expert demonstrations, it often causes overconfidence and reduces generation diversity, leaving RL with a narrowed solution space to explore. Adding entropy regularization during SFT is not a cure-all; it tends to flatten token distributions toward uniformity, increasing entropy without improving meaningful exploration capability. In this paper, we propose CurioSFT, an entropy-preserving SFT method desi"},"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":"2602.02244","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T15:53:55Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d2b46bfd5c723c037c838aa073af8a12b7faa7f6844d7fd09b33eb8e5a75d6f8","abstract_canon_sha256":"b0edf78ea82d42fbcec53138fedfb09279d97a90937ee087e14b0548d242ff3c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-15T01:21:49.972152Z","signature_b64":"cPmvlVb3LEmCxNsSivT06pTuz86j0ArkH92ky/HRazqSB/ZcOHCUwwKOLz6WTsdYcsyTC8rml4izQ5abMYp4Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"729725d67ca9e3d3b30a4ab70cde78afabe1291136e70c0098279d8872e1f2c7","last_reissued_at":"2026-07-15T01:21:49.971298Z","signature_status":"signed_v1","first_computed_at":"2026-07-15T01:21:49.971298Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Entropy-Preserving Supervised Fine-Tuning via Adaptive Self-Distillation for Large Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Dapeng Wu, Hao Gu, Hao Wang, Hongming Piao, Kaixiong Gong, Sirui Han, Xiangyu Yue, Yike Guo, Yuxiao Ye","submitted_at":"2026-02-02T15:53:55Z","abstract_excerpt":"The standard post-training recipe for large reasoning models, supervised fine-tuning followed by reinforcement learning (SFT-then-RL), may limit the benefits of the RL stage: while SFT imitates expert demonstrations, it often causes overconfidence and reduces generation diversity, leaving RL with a narrowed solution space to explore. Adding entropy regularization during SFT is not a cure-all; it tends to flatten token distributions toward uniformity, increasing entropy without improving meaningful exploration capability. In this paper, we propose CurioSFT, an entropy-preserving SFT method desi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.02244","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/2602.02244/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":"2602.02244","created_at":"2026-07-15T01:21:49.971708+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.02244v3","created_at":"2026-07-15T01:21:49.971708+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.02244","created_at":"2026-07-15T01:21:49.971708+00:00"},{"alias_kind":"pith_short_12","alias_value":"OKLSLVT4VHR5","created_at":"2026-07-15T01:21:49.971708+00:00"},{"alias_kind":"pith_short_16","alias_value":"OKLSLVT4VHR5HMYK","created_at":"2026-07-15T01:21:49.971708+00:00"},{"alias_kind":"pith_short_8","alias_value":"OKLSLVT4","created_at":"2026-07-15T01:21:49.971708+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2606.09932","citing_title":"When RL Fails after SFT: Rejuvenating Model Plasticity for Robust SFT-to-RL Handoff","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2606.04547","citing_title":"Beyond Retrieval: Learning Compact User Representations for Scalable LLM Personalization","ref_index":118,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00642","citing_title":"Learn where to Click from Yourself: On-Policy Self-Distillation for GUI Grounding","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00642","citing_title":"Learn where to Click from Yourself: On-Policy Self-Distillation for GUI Grounding","ref_index":35,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6","json":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6.json","graph_json":"https://pith.science/api/pith-number/OKLSLVT4VHR5HMYKJK3QZXTYV6/graph.json","events_json":"https://pith.science/api/pith-number/OKLSLVT4VHR5HMYKJK3QZXTYV6/events.json","paper":"https://pith.science/paper/OKLSLVT4"},"agent_actions":{"view_html":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6","download_json":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6.json","view_paper":"https://pith.science/paper/OKLSLVT4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.02244&json=true","fetch_graph":"https://pith.science/api/pith-number/OKLSLVT4VHR5HMYKJK3QZXTYV6/graph.json","fetch_events":"https://pith.science/api/pith-number/OKLSLVT4VHR5HMYKJK3QZXTYV6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6/action/storage_attestation","attest_author":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6/action/author_attestation","sign_citation":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6/action/citation_signature","submit_replication":"https://pith.science/pith/OKLSLVT4VHR5HMYKJK3QZXTYV6/action/replication_record"}},"created_at":"2026-07-15T01:21:49.971708+00:00","updated_at":"2026-07-15T01:21:49.971708+00:00"}