{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NG7FBDOSXBEX6QYMQUQPFFJ6XA","short_pith_number":"pith:NG7FBDOS","schema_version":"1.0","canonical_sha256":"69be508dd2b8497f430c8520f2953eb838b95aea2f271644dea7ad5cec2258cc","source":{"kind":"arxiv","id":"2604.14084","version":3},"attestation_state":"computed","paper":{"title":"TIP: Token Importance in On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Informative tokens in on-policy distillation come from high student entropy positions and low-entropy positions with high teacher divergence where the student is overconfident and wrong.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alborz Geramifard, Hejian Sang, Ran He, Yuanda Xu, Zhengze Zhou, Zhipeng Wang","submitted_at":"2026-04-15T16:58:24Z","abstract_excerpt":"On-policy knowledge distillation (OPD) trains a student on its own rollouts under token-level supervision from a teacher. Not all token positions matter equally, but existing views of token importance are incomplete. We ask a direct question: which tokens carry the most useful learning signal in OPD? Our answer is that informative tokens come from two regions: positions with high student entropy, and positions with low student entropy plus high teacher--student divergence, where the student is overconfident and wrong.\n  Empirically, student entropy is a strong first-order proxy: retaining $50\\"},"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":"2604.14084","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-15T16:58:24Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"1772ecceee47108b85c5ccaa6091b72facab7f9bcc774cc248b1034a9882d7ff","abstract_canon_sha256":"0beca53e5358430a4cdb5dad9dd376bdad5df3c93db884986e63c800ef1aa22e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:05:19.099085Z","signature_b64":"Bpn/KQuL19zkMwkBziVY9LEUXxz6kbhaiIKh8WzItu43cHPEfTFjuDsSqs9Gl5M5HF9ZU9eOKSzTqEieUs+nCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"69be508dd2b8497f430c8520f2953eb838b95aea2f271644dea7ad5cec2258cc","last_reissued_at":"2026-05-21T01:05:19.098471Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:05:19.098471Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TIP: Token Importance in On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Informative tokens in on-policy distillation come from high student entropy positions and low-entropy positions with high teacher divergence where the student is overconfident and wrong.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alborz Geramifard, Hejian Sang, Ran He, Yuanda Xu, Zhengze Zhou, Zhipeng Wang","submitted_at":"2026-04-15T16:58:24Z","abstract_excerpt":"On-policy knowledge distillation (OPD) trains a student on its own rollouts under token-level supervision from a teacher. Not all token positions matter equally, but existing views of token importance are incomplete. We ask a direct question: which tokens carry the most useful learning signal in OPD? Our answer is that informative tokens come from two regions: positions with high student entropy, and positions with low student entropy plus high teacher--student divergence, where the student is overconfident and wrong.\n  Empirically, student entropy is a strong first-order proxy: retaining $50\\"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"informative tokens come from two regions: positions with high student entropy, and positions with low student entropy plus high teacher--student divergence, where the student is overconfident and wrong.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That teacher-student divergence reliably flags cases where the student is factually wrong rather than simply differing in style or valid alternative answers.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"In on-policy distillation, tokens with high student entropy or low entropy plus high teacher divergence provide dense corrective signal, allowing effective training on under 20% of tokens across math and planning tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Informative tokens in on-policy distillation come from high student entropy positions and low-entropy positions with high teacher divergence where the student is overconfident and wrong.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d54ce8333edf6342f15f10dae62c2e9c86bcd444aed6d25145f84fa7b523d04b"},"source":{"id":"2604.14084","kind":"arxiv","version":3},"verdict":{"id":"14de82d4-9b9c-440d-a81f-074f75090484","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T13:23:43.447420Z","strongest_claim":"informative tokens come from two regions: positions with high student entropy, and positions with low student entropy plus high teacher--student divergence, where the student is overconfident and wrong.","one_line_summary":"In on-policy distillation, tokens with high student entropy or low entropy plus high teacher divergence provide dense corrective signal, allowing effective training on under 20% of tokens across math and planning tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That teacher-student divergence reliably flags cases where the student is factually wrong rather than simply differing in style or valid alternative answers.","pith_extraction_headline":"Informative tokens in on-policy distillation come from high student entropy positions and low-entropy positions with high teacher divergence where the student is overconfident and wrong."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.14084/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":"2604.14084","created_at":"2026-05-21T01:05:19.098559+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.14084v3","created_at":"2026-05-21T01:05:19.098559+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.14084","created_at":"2026-05-21T01:05:19.098559+00:00"},{"alias_kind":"pith_short_12","alias_value":"NG7FBDOSXBEX","created_at":"2026-05-21T01:05:19.098559+00:00"},{"alias_kind":"pith_short_16","alias_value":"NG7FBDOSXBEX6QYM","created_at":"2026-05-21T01:05:19.098559+00:00"},{"alias_kind":"pith_short_8","alias_value":"NG7FBDOS","created_at":"2026-05-21T01:05:19.098559+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":9,"sample":[{"citing_arxiv_id":"2605.07711","citing_title":"SimCT: Recovering Lost Supervision for Cross-Tokenizer On-Policy Distillation","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11739","citing_title":"Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11739","citing_title":"Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation","ref_index":104,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11739","citing_title":"Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation","ref_index":104,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08737","citing_title":"The Extrapolation Cliff in On-Policy Distillation of Near-Deterministic Structured Outputs","ref_index":42,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10194","citing_title":"TRACE: Distilling Where It Matters via Token-Routed Self On-Policy Alignment","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07396","citing_title":"Rubric-based On-policy Distillation","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07725","citing_title":"SOD: Step-wise On-policy Distillation for Small Language Model Agents","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07711","citing_title":"SimCT: Recovering Lost Supervision for Cross-Tokenizer On-Policy Distillation","ref_index":39,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA","json":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA.json","graph_json":"https://pith.science/api/pith-number/NG7FBDOSXBEX6QYMQUQPFFJ6XA/graph.json","events_json":"https://pith.science/api/pith-number/NG7FBDOSXBEX6QYMQUQPFFJ6XA/events.json","paper":"https://pith.science/paper/NG7FBDOS"},"agent_actions":{"view_html":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA","download_json":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA.json","view_paper":"https://pith.science/paper/NG7FBDOS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.14084&json=true","fetch_graph":"https://pith.science/api/pith-number/NG7FBDOSXBEX6QYMQUQPFFJ6XA/graph.json","fetch_events":"https://pith.science/api/pith-number/NG7FBDOSXBEX6QYMQUQPFFJ6XA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA/action/storage_attestation","attest_author":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA/action/author_attestation","sign_citation":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA/action/citation_signature","submit_replication":"https://pith.science/pith/NG7FBDOSXBEX6QYMQUQPFFJ6XA/action/replication_record"}},"created_at":"2026-05-21T01:05:19.098559+00:00","updated_at":"2026-05-21T01:05:19.098559+00:00"}