{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PBNEBYWPMSLX6CLPF476XANXS5","short_pith_number":"pith:PBNEBYWP","schema_version":"1.0","canonical_sha256":"785a40e2cf64977f096f2f3feb81b7977f647b2244a1fe925e4aa6d1caf252c1","source":{"kind":"arxiv","id":"2605.22731","version":1},"attestation_state":"computed","paper":{"title":"Post-Training is About States, Not Tokens: A State Distribution View of SFT, RL, and On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dong Nie","submitted_at":"2026-05-21T17:03:22Z","abstract_excerpt":"Large language model post-training methods such as supervised fine-tuning (SFT), reinforcement learning (RL), and distillation are often analyzed through their loss functions: maximum likelihood, policy gradients, forward KL, reverse KL, or related objective-level variants. We study a complementary factor: the state distribution on which supervision is applied. For an autoregressive policy, a state is a prompt plus generated prefix. SFT trains on fixed dataset states, while RL and on-policy distillation (OPD) train on states induced by the current learner. We formalize post-training as state-d"},"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":"2605.22731","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-21T17:03:22Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"45dbced9692e7ff704acf77c8b0dc4c121d1c450fa1e185bd974aa4a934e266e","abstract_canon_sha256":"89596fb6158cd2a271a48191ed152a4f6153d9db1f8e2fd9f13380fbfb3017b4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T02:04:52.347226Z","signature_b64":"XKxfrKjHhR1EHyWkZ5g4gkwk4EnSPKPuw9O5AHX7f0FRrxfNB+TtHCFvc/03E9WkqD+UdQ+umxAzkZWkiroHDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"785a40e2cf64977f096f2f3feb81b7977f647b2244a1fe925e4aa6d1caf252c1","last_reissued_at":"2026-05-22T02:04:52.346443Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T02:04:52.346443Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Post-Training is About States, Not Tokens: A State Distribution View of SFT, RL, and On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dong Nie","submitted_at":"2026-05-21T17:03:22Z","abstract_excerpt":"Large language model post-training methods such as supervised fine-tuning (SFT), reinforcement learning (RL), and distillation are often analyzed through their loss functions: maximum likelihood, policy gradients, forward KL, reverse KL, or related objective-level variants. We study a complementary factor: the state distribution on which supervision is applied. For an autoregressive policy, a state is a prompt plus generated prefix. SFT trains on fixed dataset states, while RL and on-policy distillation (OPD) train on states induced by the current learner. We formalize post-training as state-d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22731","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/2605.22731/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":"2605.22731","created_at":"2026-05-22T02:04:52.346573+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22731v1","created_at":"2026-05-22T02:04:52.346573+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22731","created_at":"2026-05-22T02:04:52.346573+00:00"},{"alias_kind":"pith_short_12","alias_value":"PBNEBYWPMSLX","created_at":"2026-05-22T02:04:52.346573+00:00"},{"alias_kind":"pith_short_16","alias_value":"PBNEBYWPMSLX6CLP","created_at":"2026-05-22T02:04:52.346573+00:00"},{"alias_kind":"pith_short_8","alias_value":"PBNEBYWP","created_at":"2026-05-22T02:04:52.346573+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/PBNEBYWPMSLX6CLPF476XANXS5","json":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5.json","graph_json":"https://pith.science/api/pith-number/PBNEBYWPMSLX6CLPF476XANXS5/graph.json","events_json":"https://pith.science/api/pith-number/PBNEBYWPMSLX6CLPF476XANXS5/events.json","paper":"https://pith.science/paper/PBNEBYWP"},"agent_actions":{"view_html":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5","download_json":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5.json","view_paper":"https://pith.science/paper/PBNEBYWP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22731&json=true","fetch_graph":"https://pith.science/api/pith-number/PBNEBYWPMSLX6CLPF476XANXS5/graph.json","fetch_events":"https://pith.science/api/pith-number/PBNEBYWPMSLX6CLPF476XANXS5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5/action/storage_attestation","attest_author":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5/action/author_attestation","sign_citation":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5/action/citation_signature","submit_replication":"https://pith.science/pith/PBNEBYWPMSLX6CLPF476XANXS5/action/replication_record"}},"created_at":"2026-05-22T02:04:52.346573+00:00","updated_at":"2026-05-22T02:04:52.346573+00:00"}