{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YJIDXWRMN7FRVEB5FGCMIH4U7I","short_pith_number":"pith:YJIDXWRM","schema_version":"1.0","canonical_sha256":"c2503bda2c6fcb1a903d2984c41f94fa263dbf3c856e769c2e3cddec6103388c","source":{"kind":"arxiv","id":"2605.17570","version":1},"attestation_state":"computed","paper":{"title":"How Off-Policy Can GRPO Be? Mu-GRPO for Efficient LLM Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chen Wei, Minghao Tian, Yunfei Xie","submitted_at":"2026-05-17T17:58:53Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has been a key driver of recent progress in reinforcement learning with verifiable rewards (RLVR) for large language models, but it is typically trained in a low-staleness, near-on-policy regime that incurs substantial system overhead. We ask a simple question: How off-policy can GRPO be? We show that GRPO-style algorithms can tolerate substantially larger rollout staleness than previously assumed, and propose Mu-GRPO, an RL training framework that organizes training into a small number (e.g., four) of large sequential generation-optimization stages. T"},"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":"2605.17570","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-17T17:58:53Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"8b5733551a143fc67f90e4a2c820e5a96b37f25c58ba9ece68b8cbb4e101999e","abstract_canon_sha256":"5010ee1055120649df60fb8ca9375faafc310860a8766a6a3e1d0788ab35daea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:46.484668Z","signature_b64":"9vJXqW61mmzd6RUP4xXFUMyVZCKrR4/yToWDvmhHPlRRsB2Yulx4O1BldAHRwlggdwnw+KvU8P+jniKlhMtGAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c2503bda2c6fcb1a903d2984c41f94fa263dbf3c856e769c2e3cddec6103388c","last_reissued_at":"2026-05-20T00:04:46.483896Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:46.483896Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Off-Policy Can GRPO Be? Mu-GRPO for Efficient LLM Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chen Wei, Minghao Tian, Yunfei Xie","submitted_at":"2026-05-17T17:58:53Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has been a key driver of recent progress in reinforcement learning with verifiable rewards (RLVR) for large language models, but it is typically trained in a low-staleness, near-on-policy regime that incurs substantial system overhead. We ask a simple question: How off-policy can GRPO be? We show that GRPO-style algorithms can tolerate substantially larger rollout staleness than previously assumed, and propose Mu-GRPO, an RL training framework that organizes training into a small number (e.g., four) of large sequential generation-optimization stages. T"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.17570","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/2605.17570/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.595403Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T21:21:57.526963Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"b5b6bb21d89c7c8a60ebfec6be41e2b186e946c1dbac402bd568662b1ca01dfe"},"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":"2605.17570","created_at":"2026-05-20T00:04:46.484010+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.17570v1","created_at":"2026-05-20T00:04:46.484010+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.17570","created_at":"2026-05-20T00:04:46.484010+00:00"},{"alias_kind":"pith_short_12","alias_value":"YJIDXWRMN7FR","created_at":"2026-05-20T00:04:46.484010+00:00"},{"alias_kind":"pith_short_16","alias_value":"YJIDXWRMN7FRVEB5","created_at":"2026-05-20T00:04:46.484010+00:00"},{"alias_kind":"pith_short_8","alias_value":"YJIDXWRM","created_at":"2026-05-20T00:04:46.484010+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/YJIDXWRMN7FRVEB5FGCMIH4U7I","json":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I.json","graph_json":"https://pith.science/api/pith-number/YJIDXWRMN7FRVEB5FGCMIH4U7I/graph.json","events_json":"https://pith.science/api/pith-number/YJIDXWRMN7FRVEB5FGCMIH4U7I/events.json","paper":"https://pith.science/paper/YJIDXWRM"},"agent_actions":{"view_html":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I","download_json":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I.json","view_paper":"https://pith.science/paper/YJIDXWRM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.17570&json=true","fetch_graph":"https://pith.science/api/pith-number/YJIDXWRMN7FRVEB5FGCMIH4U7I/graph.json","fetch_events":"https://pith.science/api/pith-number/YJIDXWRMN7FRVEB5FGCMIH4U7I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I/action/storage_attestation","attest_author":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I/action/author_attestation","sign_citation":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I/action/citation_signature","submit_replication":"https://pith.science/pith/YJIDXWRMN7FRVEB5FGCMIH4U7I/action/replication_record"}},"created_at":"2026-05-20T00:04:46.484010+00:00","updated_at":"2026-05-20T00:04:46.484010+00:00"}