{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:C3NWHLP7UCUYPUDA6KXGHSXBIQ","short_pith_number":"pith:C3NWHLP7","schema_version":"1.0","canonical_sha256":"16db63adffa0a987d060f2ae63cae1440d02ebe7bc6b2e52419340d0657eb544","source":{"kind":"arxiv","id":"2506.05433","version":1},"attestation_state":"computed","paper":{"title":"Prefix Grouper: Efficient GRPO Training through Shared-Prefix Forward","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jing Liu, Junxian Cai, Longteng Guo, Qingbin Liu, Tongtian Yue, Xi Chen, Yepeng Tang, Zikang Liu","submitted_at":"2025-06-05T09:13:37Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) enhances policy learning by computing gradients from relative comparisons among candidate outputs that share a common input prefix. Despite its effectiveness, GRPO introduces substantial computational overhead when processing long shared prefixes, which must be redundantly encoded for each group member. This inefficiency becomes a major scalability bottleneck in long-context learning scenarios. We propose Prefix Grouper, an efficient GRPO training algorithm that eliminates redundant prefix computation via a Shared-Prefix Forward strategy. In particular"},"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":"2506.05433","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-05T09:13:37Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"ff995b34a4a2294f7d2e2808adb86553a23ca17d150dfe000f3bdc2c87371681","abstract_canon_sha256":"a1043a179a98c9a4de85d25ca30b7fa04540e7f98ff61de568bf5609318aa03d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:17:02.597978Z","signature_b64":"+LQM7MLCCN8Di1EB8ClDL0CqSzTehpwbZ+wXKHUYBaJxR46IxClNKYD+DsMPQq4rtQn/TBamYo7nZIkzjSA7AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"16db63adffa0a987d060f2ae63cae1440d02ebe7bc6b2e52419340d0657eb544","last_reissued_at":"2026-07-05T11:17:02.597500Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:17:02.597500Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Prefix Grouper: Efficient GRPO Training through Shared-Prefix Forward","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jing Liu, Junxian Cai, Longteng Guo, Qingbin Liu, Tongtian Yue, Xi Chen, Yepeng Tang, Zikang Liu","submitted_at":"2025-06-05T09:13:37Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) enhances policy learning by computing gradients from relative comparisons among candidate outputs that share a common input prefix. Despite its effectiveness, GRPO introduces substantial computational overhead when processing long shared prefixes, which must be redundantly encoded for each group member. This inefficiency becomes a major scalability bottleneck in long-context learning scenarios. We propose Prefix Grouper, an efficient GRPO training algorithm that eliminates redundant prefix computation via a Shared-Prefix Forward strategy. In particular"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.05433","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/2506.05433/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":"2506.05433","created_at":"2026-07-05T11:17:02.597552+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.05433v1","created_at":"2026-07-05T11:17:02.597552+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.05433","created_at":"2026-07-05T11:17:02.597552+00:00"},{"alias_kind":"pith_short_12","alias_value":"C3NWHLP7UCUY","created_at":"2026-07-05T11:17:02.597552+00:00"},{"alias_kind":"pith_short_16","alias_value":"C3NWHLP7UCUYPUDA","created_at":"2026-07-05T11:17:02.597552+00:00"},{"alias_kind":"pith_short_8","alias_value":"C3NWHLP7","created_at":"2026-07-05T11:17:02.597552+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.02218","citing_title":"Faster Synchronous On-Policy RL via Straggler-Aware Group Sizing","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01143","citing_title":"Schedule-Level Shared-Prefix Reuse for LLM RL Training","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15422","citing_title":"DualKV: Shared-Prompt Flash Attention for Efficient RL Training with Large Rollouts and Long Contexts","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15422","citing_title":"DualKV: Shared-Prompt Flash Attention for Efficient RL Training with Large Rollouts and Long Contexts","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2509.10026","citing_title":"LaV-CoT: Language-Aware Visual CoT with Multi-Aspect Reward Optimization for Real-World Multilingual VQA","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2511.00413","citing_title":"Tree Training: Accelerating Agentic LLMs Training via Shared Prefix Reuse","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ","json":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ.json","graph_json":"https://pith.science/api/pith-number/C3NWHLP7UCUYPUDA6KXGHSXBIQ/graph.json","events_json":"https://pith.science/api/pith-number/C3NWHLP7UCUYPUDA6KXGHSXBIQ/events.json","paper":"https://pith.science/paper/C3NWHLP7"},"agent_actions":{"view_html":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ","download_json":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ.json","view_paper":"https://pith.science/paper/C3NWHLP7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.05433&json=true","fetch_graph":"https://pith.science/api/pith-number/C3NWHLP7UCUYPUDA6KXGHSXBIQ/graph.json","fetch_events":"https://pith.science/api/pith-number/C3NWHLP7UCUYPUDA6KXGHSXBIQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ/action/storage_attestation","attest_author":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ/action/author_attestation","sign_citation":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ/action/citation_signature","submit_replication":"https://pith.science/pith/C3NWHLP7UCUYPUDA6KXGHSXBIQ/action/replication_record"}},"created_at":"2026-07-05T11:17:02.597552+00:00","updated_at":"2026-07-05T11:17:02.597552+00:00"}