{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:U5GXUTDM33C2YL3KFARH32XJFR","short_pith_number":"pith:U5GXUTDM","schema_version":"1.0","canonical_sha256":"a74d7a4c6cdec5ac2f6a28227deae92c545de5fdc76e38e01cb04edcba532430","source":{"kind":"arxiv","id":"2506.04746","version":1},"attestation_state":"computed","paper":{"title":"Multi-Layer GRPO: Enhancing Reasoning and Self-Correction in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Baiqiao Wang, Fei Ding, youwei wang, Zijian Zeng","submitted_at":"2025-06-05T08:27:34Z","abstract_excerpt":"The Group Relative Policy Optimization (GRPO) algorithm has demonstrated considerable success in enhancing the reasoning capabilities of large language models (LLMs), as evidenced by DeepSeek-R1. However, the absence of intermediate supervision in GRPO frequently leads to inefficient exploration dynamics. A single error in a complex reasoning chain can invalidate the entire solution, resulting in abrupt reward vanishing and compromising training stability.To address these challenges, we propose MGRPO (Multi-layer GRPO). MGRPO operates in two layers: the first layer employs standard GRPO to gen"},"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":"2506.04746","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-05T08:27:34Z","cross_cats_sorted":[],"title_canon_sha256":"5fd3876233f5714091a8125f44f4f690f1f84e9e0742f1857a75b9712f8d1148","abstract_canon_sha256":"1c6e6620c1ff6e52cf1bb3893f133d4b8d2d3e0d70d67085602c21d7686b8f0a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:19.668805Z","signature_b64":"dVlueJWY3JEADJicPUuN8VfgOIJvAvj+bHtc3ZexBn7ncN6z5DXW94ev/IwfrDThRHWLWToHkB2krB+IeD6PCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a74d7a4c6cdec5ac2f6a28227deae92c545de5fdc76e38e01cb04edcba532430","last_reissued_at":"2026-07-05T11:16:19.668358Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:19.668358Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Layer GRPO: Enhancing Reasoning and Self-Correction in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Baiqiao Wang, Fei Ding, youwei wang, Zijian Zeng","submitted_at":"2025-06-05T08:27:34Z","abstract_excerpt":"The Group Relative Policy Optimization (GRPO) algorithm has demonstrated considerable success in enhancing the reasoning capabilities of large language models (LLMs), as evidenced by DeepSeek-R1. However, the absence of intermediate supervision in GRPO frequently leads to inefficient exploration dynamics. A single error in a complex reasoning chain can invalidate the entire solution, resulting in abrupt reward vanishing and compromising training stability.To address these challenges, we propose MGRPO (Multi-layer GRPO). MGRPO operates in two layers: the first layer employs standard GRPO to gen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.04746","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/2506.04746/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":"2506.04746","created_at":"2026-07-05T11:16:19.668419+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.04746v1","created_at":"2026-07-05T11:16:19.668419+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.04746","created_at":"2026-07-05T11:16:19.668419+00:00"},{"alias_kind":"pith_short_12","alias_value":"U5GXUTDM33C2","created_at":"2026-07-05T11:16:19.668419+00:00"},{"alias_kind":"pith_short_16","alias_value":"U5GXUTDM33C2YL3K","created_at":"2026-07-05T11:16:19.668419+00:00"},{"alias_kind":"pith_short_8","alias_value":"U5GXUTDM","created_at":"2026-07-05T11:16:19.668419+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/U5GXUTDM33C2YL3KFARH32XJFR","json":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR.json","graph_json":"https://pith.science/api/pith-number/U5GXUTDM33C2YL3KFARH32XJFR/graph.json","events_json":"https://pith.science/api/pith-number/U5GXUTDM33C2YL3KFARH32XJFR/events.json","paper":"https://pith.science/paper/U5GXUTDM"},"agent_actions":{"view_html":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR","download_json":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR.json","view_paper":"https://pith.science/paper/U5GXUTDM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.04746&json=true","fetch_graph":"https://pith.science/api/pith-number/U5GXUTDM33C2YL3KFARH32XJFR/graph.json","fetch_events":"https://pith.science/api/pith-number/U5GXUTDM33C2YL3KFARH32XJFR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR/action/storage_attestation","attest_author":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR/action/author_attestation","sign_citation":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR/action/citation_signature","submit_replication":"https://pith.science/pith/U5GXUTDM33C2YL3KFARH32XJFR/action/replication_record"}},"created_at":"2026-07-05T11:16:19.668419+00:00","updated_at":"2026-07-05T11:16:19.668419+00:00"}