{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:EFLZ7JRHWCRT4FLCHOJA7KBDLA","short_pith_number":"pith:EFLZ7JRH","schema_version":"1.0","canonical_sha256":"21579fa627b0a33e15623b920fa823580707944e14085644c26bc840bb3c4e92","source":{"kind":"arxiv","id":"2606.08779","version":1},"attestation_state":"computed","paper":{"title":"Reformulate LLM Reinforcement Learning for Efficient Training under Black-box Discrepancy","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hongyao Tang, Jiashun Liu, Jing Liang, Ling Pan, Runze Liu, Xu Wan","submitted_at":"2026-06-07T18:49:33Z","abstract_excerpt":"Reinforcement Learning (RL) has emerged as a pivotal post-training paradigm, yet it frequently suffers from unpredictable sub-optimum performance or even training collapses. Recent findings attribute these failures to a hidden train-inference discrepancy (or mismatch), stemming from the disparate underlying engines and architecture. We find that the training policy can actively self-correct such a discrepancy when provided with an appropriate learning signal. Then, we further empirically identify a discrepancy tolerance region: within this region, aggressively narrowing the discrepancy can sup"},"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":"2606.08779","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-07T18:49:33Z","cross_cats_sorted":[],"title_canon_sha256":"606b8dfbfdbca28f7233297ab6b62a10d63420e8c12339e0b78370fdc86c44f2","abstract_canon_sha256":"4596793964a94577ade9323ca86155660ce8b82d4941c826648aca9bad6f3e9b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T02:07:38.981158Z","signature_b64":"DFLJUXqB3rSUzl7AbBDDYlRENSD5QBLJ+OVbDNNaTt621+dzquCh+rxmULWgdV3YFo5zeupJjRU94PjmkDs6Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"21579fa627b0a33e15623b920fa823580707944e14085644c26bc840bb3c4e92","last_reissued_at":"2026-06-09T02:07:38.980320Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T02:07:38.980320Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reformulate LLM Reinforcement Learning for Efficient Training under Black-box Discrepancy","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hongyao Tang, Jiashun Liu, Jing Liang, Ling Pan, Runze Liu, Xu Wan","submitted_at":"2026-06-07T18:49:33Z","abstract_excerpt":"Reinforcement Learning (RL) has emerged as a pivotal post-training paradigm, yet it frequently suffers from unpredictable sub-optimum performance or even training collapses. Recent findings attribute these failures to a hidden train-inference discrepancy (or mismatch), stemming from the disparate underlying engines and architecture. We find that the training policy can actively self-correct such a discrepancy when provided with an appropriate learning signal. Then, we further empirically identify a discrepancy tolerance region: within this region, aggressively narrowing the discrepancy can sup"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.08779","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/2606.08779/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":"2606.08779","created_at":"2026-06-09T02:07:38.980659+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.08779v1","created_at":"2026-06-09T02:07:38.980659+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.08779","created_at":"2026-06-09T02:07:38.980659+00:00"},{"alias_kind":"pith_short_12","alias_value":"EFLZ7JRHWCRT","created_at":"2026-06-09T02:07:38.980659+00:00"},{"alias_kind":"pith_short_16","alias_value":"EFLZ7JRHWCRT4FLC","created_at":"2026-06-09T02:07:38.980659+00:00"},{"alias_kind":"pith_short_8","alias_value":"EFLZ7JRH","created_at":"2026-06-09T02:07:38.980659+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/EFLZ7JRHWCRT4FLCHOJA7KBDLA","json":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA.json","graph_json":"https://pith.science/api/pith-number/EFLZ7JRHWCRT4FLCHOJA7KBDLA/graph.json","events_json":"https://pith.science/api/pith-number/EFLZ7JRHWCRT4FLCHOJA7KBDLA/events.json","paper":"https://pith.science/paper/EFLZ7JRH"},"agent_actions":{"view_html":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA","download_json":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA.json","view_paper":"https://pith.science/paper/EFLZ7JRH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.08779&json=true","fetch_graph":"https://pith.science/api/pith-number/EFLZ7JRHWCRT4FLCHOJA7KBDLA/graph.json","fetch_events":"https://pith.science/api/pith-number/EFLZ7JRHWCRT4FLCHOJA7KBDLA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA/action/storage_attestation","attest_author":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA/action/author_attestation","sign_citation":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA/action/citation_signature","submit_replication":"https://pith.science/pith/EFLZ7JRHWCRT4FLCHOJA7KBDLA/action/replication_record"}},"created_at":"2026-06-09T02:07:38.980659+00:00","updated_at":"2026-06-09T02:07:38.980659+00:00"}