{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AKJYG3H5L5R54H7NNH7ADTNAP7","short_pith_number":"pith:AKJYG3H5","schema_version":"1.0","canonical_sha256":"0293836cfd5f63de1fed69fe01cda07ff03bbee3103f251e2bb85663643cd5f9","source":{"kind":"arxiv","id":"2606.13657","version":1},"attestation_state":"computed","paper":{"title":"Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Guo Yu, Han-Jia Ye, Hao-Xuan Ma, Jun-Peng Jiang, Wenlin Liu, Yulan Hu","submitted_at":"2026-06-11T17:54:09Z","abstract_excerpt":"On-policy distillation (\\textsc{OPD}) has recently become a prominent post-training recipe as it combines two desirable ingredients: on-policy student trajectories and dense teacher supervision, yet how this hybrid changes a model's parameters remains unclear. Across several language and vision-language model pairs and use cases, our analysis yields two main findings. On sparsity, \\textsc{OPD}-style updates are small and coordinate-sparse. They are distributed across layers and are usually FFN-heavy. This sparse structure is operationally useful: training only the discovered subnetwork recover"},"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":"2606.13657","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-11T17:54:09Z","cross_cats_sorted":[],"title_canon_sha256":"8e8200d97f5be28770698b005885868b4ca6d5f266ba9f0648be93d5ca299a38","abstract_canon_sha256":"c7dd660e027bda0431c5712ab2183104fb4bd8b263f5cffd87912d7a82ef5821"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-12T01:10:21.490188Z","signature_b64":"QvYH38N+YcJFWFLwxmEzTLNHevZDmDv6MMrGLe2bIbAIOWDjDei9uozTgsW9rF1N5goquo+434IPwvzppdvEDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0293836cfd5f63de1fed69fe01cda07ff03bbee3103f251e2bb85663643cd5f9","last_reissued_at":"2026-06-12T01:10:21.489301Z","signature_status":"signed_v1","first_computed_at":"2026-06-12T01:10:21.489301Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Guo Yu, Han-Jia Ye, Hao-Xuan Ma, Jun-Peng Jiang, Wenlin Liu, Yulan Hu","submitted_at":"2026-06-11T17:54:09Z","abstract_excerpt":"On-policy distillation (\\textsc{OPD}) has recently become a prominent post-training recipe as it combines two desirable ingredients: on-policy student trajectories and dense teacher supervision, yet how this hybrid changes a model's parameters remains unclear. Across several language and vision-language model pairs and use cases, our analysis yields two main findings. On sparsity, \\textsc{OPD}-style updates are small and coordinate-sparse. They are distributed across layers and are usually FFN-heavy. This sparse structure is operationally useful: training only the discovered subnetwork recover"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.13657","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/2606.13657/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":"2606.13657","created_at":"2026-06-12T01:10:21.489454+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.13657v1","created_at":"2026-06-12T01:10:21.489454+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.13657","created_at":"2026-06-12T01:10:21.489454+00:00"},{"alias_kind":"pith_short_12","alias_value":"AKJYG3H5L5R5","created_at":"2026-06-12T01:10:21.489454+00:00"},{"alias_kind":"pith_short_16","alias_value":"AKJYG3H5L5R54H7N","created_at":"2026-06-12T01:10:21.489454+00:00"},{"alias_kind":"pith_short_8","alias_value":"AKJYG3H5","created_at":"2026-06-12T01:10:21.489454+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/AKJYG3H5L5R54H7NNH7ADTNAP7","json":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7.json","graph_json":"https://pith.science/api/pith-number/AKJYG3H5L5R54H7NNH7ADTNAP7/graph.json","events_json":"https://pith.science/api/pith-number/AKJYG3H5L5R54H7NNH7ADTNAP7/events.json","paper":"https://pith.science/paper/AKJYG3H5"},"agent_actions":{"view_html":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7","download_json":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7.json","view_paper":"https://pith.science/paper/AKJYG3H5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.13657&json=true","fetch_graph":"https://pith.science/api/pith-number/AKJYG3H5L5R54H7NNH7ADTNAP7/graph.json","fetch_events":"https://pith.science/api/pith-number/AKJYG3H5L5R54H7NNH7ADTNAP7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7/action/storage_attestation","attest_author":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7/action/author_attestation","sign_citation":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7/action/citation_signature","submit_replication":"https://pith.science/pith/AKJYG3H5L5R54H7NNH7ADTNAP7/action/replication_record"}},"created_at":"2026-06-12T01:10:21.489454+00:00","updated_at":"2026-06-12T01:10:21.489454+00:00"}