{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XQW5AIMXSI4WCRD2KM4FR3TNML","short_pith_number":"pith:XQW5AIMX","schema_version":"1.0","canonical_sha256":"bc2dd02197923961447a533858ee6d62f2a510e3d1af153e3b2094288f4eb4c3","source":{"kind":"arxiv","id":"2606.09821","version":1},"attestation_state":"computed","paper":{"title":"Rethinking the Divergence Regularization in LLM RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jiarui Yao, Liefeng Bo, Penghui Qi, Tianyu Pang, Wee Sun Lee, Xiangxin Zhou","submitted_at":"2026-06-08T17:58:23Z","abstract_excerpt":"Reinforcement learning (RL) has become a key component of post-training large language models (LLMs). In practice, LLM RL is often off-policy because of training-inference mismatch and policy staleness, making trust-region control essential for stable optimization. Mainstream methods such as PPO and GRPO approximate this control with a ratio-clipping mechanism, but the importance ratio can be a poor proxy for distributional shift in long-tailed vocabularies. Recent work such as DPPO addresses this mismatch by replacing ratio-based clipping with a divergence-based mask, yielding a trust region "},"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":"2606.09821","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-08T17:58:23Z","cross_cats_sorted":[],"title_canon_sha256":"dd093fc1d6cd7ab56c3ea1905269a61b09d5767173dd3bb52397cf4c36116d72","abstract_canon_sha256":"82d9ffc9ac850e3b1a4722068eb2441c2768b63a9b0eb5ece39df29de369c1ba"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T02:09:12.426206Z","signature_b64":"is771FhLFx2XXczGhkR5/njA6+5VyakNBNL4F8gD5DBk826uDOs5+G4fL0IccWB9yYAzbXKY2bc7ZxUnsyHcCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bc2dd02197923961447a533858ee6d62f2a510e3d1af153e3b2094288f4eb4c3","last_reissued_at":"2026-06-09T02:09:12.425341Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T02:09:12.425341Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking the Divergence Regularization in LLM RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jiarui Yao, Liefeng Bo, Penghui Qi, Tianyu Pang, Wee Sun Lee, Xiangxin Zhou","submitted_at":"2026-06-08T17:58:23Z","abstract_excerpt":"Reinforcement learning (RL) has become a key component of post-training large language models (LLMs). In practice, LLM RL is often off-policy because of training-inference mismatch and policy staleness, making trust-region control essential for stable optimization. Mainstream methods such as PPO and GRPO approximate this control with a ratio-clipping mechanism, but the importance ratio can be a poor proxy for distributional shift in long-tailed vocabularies. Recent work such as DPPO addresses this mismatch by replacing ratio-based clipping with a divergence-based mask, yielding a trust region "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.09821","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/2606.09821/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":"2606.09821","created_at":"2026-06-09T02:09:12.425490+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.09821v1","created_at":"2026-06-09T02:09:12.425490+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.09821","created_at":"2026-06-09T02:09:12.425490+00:00"},{"alias_kind":"pith_short_12","alias_value":"XQW5AIMXSI4W","created_at":"2026-06-09T02:09:12.425490+00:00"},{"alias_kind":"pith_short_16","alias_value":"XQW5AIMXSI4WCRD2","created_at":"2026-06-09T02:09:12.425490+00:00"},{"alias_kind":"pith_short_8","alias_value":"XQW5AIMX","created_at":"2026-06-09T02:09:12.425490+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/XQW5AIMXSI4WCRD2KM4FR3TNML","json":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML.json","graph_json":"https://pith.science/api/pith-number/XQW5AIMXSI4WCRD2KM4FR3TNML/graph.json","events_json":"https://pith.science/api/pith-number/XQW5AIMXSI4WCRD2KM4FR3TNML/events.json","paper":"https://pith.science/paper/XQW5AIMX"},"agent_actions":{"view_html":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML","download_json":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML.json","view_paper":"https://pith.science/paper/XQW5AIMX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.09821&json=true","fetch_graph":"https://pith.science/api/pith-number/XQW5AIMXSI4WCRD2KM4FR3TNML/graph.json","fetch_events":"https://pith.science/api/pith-number/XQW5AIMXSI4WCRD2KM4FR3TNML/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML/action/storage_attestation","attest_author":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML/action/author_attestation","sign_citation":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML/action/citation_signature","submit_replication":"https://pith.science/pith/XQW5AIMXSI4WCRD2KM4FR3TNML/action/replication_record"}},"created_at":"2026-06-09T02:09:12.425490+00:00","updated_at":"2026-06-09T02:09:12.425490+00:00"}