{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:W23KVJ3SVBYLN66OVRWPOTTQAG","short_pith_number":"pith:W23KVJ3S","canonical_record":{"source":{"id":"2505.12504","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-18T17:44:53Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"bcf19939b4bee633713401439962696039d09aeffd5349a13a436ed5e248e7de","abstract_canon_sha256":"f24b1d6236949b52555eaeef6e460349d3731e6e710c19bf2a36611eb730e49c"},"schema_version":"1.0"},"canonical_sha256":"b6b6aaa772a870b6fbceac6cf74e7001935b9e94ffe38b89df53c8feec19bdb6","source":{"kind":"arxiv","id":"2505.12504","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.12504","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"arxiv_version","alias_value":"2505.12504v1","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.12504","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"pith_short_12","alias_value":"W23KVJ3SVBYL","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"pith_short_16","alias_value":"W23KVJ3SVBYLN66O","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"pith_short_8","alias_value":"W23KVJ3S","created_at":"2026-07-05T11:05:05Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:W23KVJ3SVBYLN66OVRWPOTTQAG","target":"record","payload":{"canonical_record":{"source":{"id":"2505.12504","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-18T17:44:53Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"bcf19939b4bee633713401439962696039d09aeffd5349a13a436ed5e248e7de","abstract_canon_sha256":"f24b1d6236949b52555eaeef6e460349d3731e6e710c19bf2a36611eb730e49c"},"schema_version":"1.0"},"canonical_sha256":"b6b6aaa772a870b6fbceac6cf74e7001935b9e94ffe38b89df53c8feec19bdb6","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:05:05.422938Z","signature_b64":"uFQdyf/XoPK2MPeG5j9PtoiLIk9DodB4+EKb75UAYmCBmwPu35kr361+UZo77e5wBHxwrsqxH/fZZTy6JyONAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b6b6aaa772a870b6fbceac6cf74e7001935b9e94ffe38b89df53c8feec19bdb6","last_reissued_at":"2026-07-05T11:05:05.422515Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:05:05.422515Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2505.12504","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:05:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"96GHeRiZZ10PdRTwqrO9fqtNW+t6CUEv/HC5wnXlAqY9YaKWDEtNaRn5ql5t0AnssLbdOhWgWAfEsYTZz0P0BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-07T04:16:44.355599Z"},"content_sha256":"599af83c2025ddd42feb437ad986adbd6a27c2391c2202772c5debc548189738","schema_version":"1.0","event_id":"sha256:599af83c2025ddd42feb437ad986adbd6a27c2391c2202772c5debc548189738"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:W23KVJ3SVBYLN66OVRWPOTTQAG","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"CPGD: Toward Stable Rule-based Reinforcement Learning for Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chao Yu, Fanqing Meng, Lingxiao Du, Qiaosheng Zhang, Wenqi Shao, Zhixiang Zhou, Zongkai Liu","submitted_at":"2025-05-18T17:44:53Z","abstract_excerpt":"Recent advances in rule-based reinforcement learning (RL) have significantly improved the reasoning capability of language models (LMs) with rule-based rewards. However, existing RL methods -- such as GRPO, REINFORCE++, and RLOO -- often suffer from training instability, where large policy updates and improper clipping can lead to training collapse. To address this issue, we propose Clipped Policy Gradient Optimization with Policy Drift (CPGD), a novel algorithm designed to stabilize policy learning in LMs. CPGD introduces a policy drift constraint based on KL divergence to dynamically regular"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.12504","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/2505.12504/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:05:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HIVFchbQhkIFU3qBgiTaJi7kUkrVzWRMhAF60drFbTAIpein873UzSPyB9TIXSrFPGlpTGgwCXoQIHD2C+dNDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-07T04:16:44.356028Z"},"content_sha256":"81e7838a0005392b0f5e2598e9c66ff0af36efee75e2b88b33de56cf1fec71a5","schema_version":"1.0","event_id":"sha256:81e7838a0005392b0f5e2598e9c66ff0af36efee75e2b88b33de56cf1fec71a5"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/W23KVJ3SVBYLN66OVRWPOTTQAG/bundle.json","state_url":"https://pith.science/pith/W23KVJ3SVBYLN66OVRWPOTTQAG/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/W23KVJ3SVBYLN66OVRWPOTTQAG/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-07T04:16:44Z","links":{"resolver":"https://pith.science/pith/W23KVJ3SVBYLN66OVRWPOTTQAG","bundle":"https://pith.science/pith/W23KVJ3SVBYLN66OVRWPOTTQAG/bundle.json","state":"https://pith.science/pith/W23KVJ3SVBYLN66OVRWPOTTQAG/state.json","well_known_bundle":"https://pith.science/.well-known/pith/W23KVJ3SVBYLN66OVRWPOTTQAG/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:W23KVJ3SVBYLN66OVRWPOTTQAG","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f24b1d6236949b52555eaeef6e460349d3731e6e710c19bf2a36611eb730e49c","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-18T17:44:53Z","title_canon_sha256":"bcf19939b4bee633713401439962696039d09aeffd5349a13a436ed5e248e7de"},"schema_version":"1.0","source":{"id":"2505.12504","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.12504","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"arxiv_version","alias_value":"2505.12504v1","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.12504","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"pith_short_12","alias_value":"W23KVJ3SVBYL","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"pith_short_16","alias_value":"W23KVJ3SVBYLN66O","created_at":"2026-07-05T11:05:05Z"},{"alias_kind":"pith_short_8","alias_value":"W23KVJ3S","created_at":"2026-07-05T11:05:05Z"}],"graph_snapshots":[{"event_id":"sha256:81e7838a0005392b0f5e2598e9c66ff0af36efee75e2b88b33de56cf1fec71a5","target":"graph","created_at":"2026-07-05T11:05:05Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2505.12504/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Recent advances in rule-based reinforcement learning (RL) have significantly improved the reasoning capability of language models (LMs) with rule-based rewards. However, existing RL methods -- such as GRPO, REINFORCE++, and RLOO -- often suffer from training instability, where large policy updates and improper clipping can lead to training collapse. To address this issue, we propose Clipped Policy Gradient Optimization with Policy Drift (CPGD), a novel algorithm designed to stabilize policy learning in LMs. CPGD introduces a policy drift constraint based on KL divergence to dynamically regular","authors_text":"Chao Yu, Fanqing Meng, Lingxiao Du, Qiaosheng Zhang, Wenqi Shao, Zhixiang Zhou, Zongkai Liu","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-18T17:44:53Z","title":"CPGD: Toward Stable Rule-based Reinforcement Learning for Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.12504","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:599af83c2025ddd42feb437ad986adbd6a27c2391c2202772c5debc548189738","target":"record","created_at":"2026-07-05T11:05:05Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f24b1d6236949b52555eaeef6e460349d3731e6e710c19bf2a36611eb730e49c","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-18T17:44:53Z","title_canon_sha256":"bcf19939b4bee633713401439962696039d09aeffd5349a13a436ed5e248e7de"},"schema_version":"1.0","source":{"id":"2505.12504","kind":"arxiv","version":1}},"canonical_sha256":"b6b6aaa772a870b6fbceac6cf74e7001935b9e94ffe38b89df53c8feec19bdb6","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"b6b6aaa772a870b6fbceac6cf74e7001935b9e94ffe38b89df53c8feec19bdb6","first_computed_at":"2026-07-05T11:05:05.422515Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:05:05.422515Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"uFQdyf/XoPK2MPeG5j9PtoiLIk9DodB4+EKb75UAYmCBmwPu35kr361+UZo77e5wBHxwrsqxH/fZZTy6JyONAQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:05:05.422938Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.12504","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:599af83c2025ddd42feb437ad986adbd6a27c2391c2202772c5debc548189738","sha256:81e7838a0005392b0f5e2598e9c66ff0af36efee75e2b88b33de56cf1fec71a5"],"state_sha256":"5752fed98038fd0184dc2e7813412d7d20f734c06b649391f124e3b17f59a9c9"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"UiQk+xCH0iFOQmBHMsgq3MLMrzFl1XpgaOkubjOByhsZvcWBanbJtzhjLwo0if+f1Nm4C25Y3xuebrxY4SA+DQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-07T04:16:44.357990Z","bundle_sha256":"2f8a361e4bf5935fef3e58b3a19d35f345ebb30662d21406991874082790687a"}}