{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ILHP74PGQZOS4SYTST5X4FFPFB","short_pith_number":"pith:ILHP74PG","schema_version":"1.0","canonical_sha256":"42cefff1e6865d2e4b1394fb7e14af284946465ef890543292792aa0c5bbd6d0","source":{"kind":"arxiv","id":"2601.09085","version":2},"attestation_state":"computed","paper":{"title":"MMR-GRPO: Accelerating GRPO-Style Training through Diversity-Aware Reward Reweighting","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.IR"],"primary_cat":"cs.LG","authors_text":"Kangda Wei, Ruihong Huang","submitted_at":"2026-01-14T02:35:19Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has become a standard approach for training mathematical reasoning models; however, its reliance on multiple completions per prompt makes training computationally expensive. Although recent work has reduced the number of training steps required to reach peak performance, the overall wall-clock training time often remains unchanged or even increases due to higher per-step cost. We propose MMR-GRPO, which integrates Maximal Marginal Relevance to reweigh rewards based on completion diversity. Our key insight is that semantically redundant completions cont"},"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":"2601.09085","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-01-14T02:35:19Z","cross_cats_sorted":["cs.AI","cs.CL","cs.IR"],"title_canon_sha256":"6f520fcecbe667c45e4c757cea65e983242b195c1fb05ecf8f84113013888846","abstract_canon_sha256":"2404a6976c56612648fbac1a774c415748f525bb59a70ab3994ea333fc14c44d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:05:12.604457Z","signature_b64":"cJ5FcDsyq1aGXcTQ0nIbEdUYln/vV4z6pe2yeqdXwX+dvjuWlgPb+VZianDwE2eYZzN/BhEihelBpgQbu0JpAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"42cefff1e6865d2e4b1394fb7e14af284946465ef890543292792aa0c5bbd6d0","last_reissued_at":"2026-06-09T01:05:12.603979Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:05:12.603979Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MMR-GRPO: Accelerating GRPO-Style Training through Diversity-Aware Reward Reweighting","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.IR"],"primary_cat":"cs.LG","authors_text":"Kangda Wei, Ruihong Huang","submitted_at":"2026-01-14T02:35:19Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has become a standard approach for training mathematical reasoning models; however, its reliance on multiple completions per prompt makes training computationally expensive. Although recent work has reduced the number of training steps required to reach peak performance, the overall wall-clock training time often remains unchanged or even increases due to higher per-step cost. We propose MMR-GRPO, which integrates Maximal Marginal Relevance to reweigh rewards based on completion diversity. Our key insight is that semantically redundant completions cont"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.09085","kind":"arxiv","version":2},"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/2601.09085/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":"2601.09085","created_at":"2026-06-09T01:05:12.604039+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.09085v2","created_at":"2026-06-09T01:05:12.604039+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.09085","created_at":"2026-06-09T01:05:12.604039+00:00"},{"alias_kind":"pith_short_12","alias_value":"ILHP74PGQZOS","created_at":"2026-06-09T01:05:12.604039+00:00"},{"alias_kind":"pith_short_16","alias_value":"ILHP74PGQZOS4SYT","created_at":"2026-06-09T01:05:12.604039+00:00"},{"alias_kind":"pith_short_8","alias_value":"ILHP74PG","created_at":"2026-06-09T01:05:12.604039+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.21801","citing_title":"Why Semantic Entropy Fails: Geometry-Aware and Calibrated Uncertainty for Policy Optimization","ref_index":60,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17333","citing_title":"Leveraging Error Diversity in Group Rollouts for Reinforcement Learning","ref_index":28,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB","json":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB.json","graph_json":"https://pith.science/api/pith-number/ILHP74PGQZOS4SYTST5X4FFPFB/graph.json","events_json":"https://pith.science/api/pith-number/ILHP74PGQZOS4SYTST5X4FFPFB/events.json","paper":"https://pith.science/paper/ILHP74PG"},"agent_actions":{"view_html":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB","download_json":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB.json","view_paper":"https://pith.science/paper/ILHP74PG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.09085&json=true","fetch_graph":"https://pith.science/api/pith-number/ILHP74PGQZOS4SYTST5X4FFPFB/graph.json","fetch_events":"https://pith.science/api/pith-number/ILHP74PGQZOS4SYTST5X4FFPFB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB/action/storage_attestation","attest_author":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB/action/author_attestation","sign_citation":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB/action/citation_signature","submit_replication":"https://pith.science/pith/ILHP74PGQZOS4SYTST5X4FFPFB/action/replication_record"}},"created_at":"2026-06-09T01:05:12.604039+00:00","updated_at":"2026-06-09T01:05:12.604039+00:00"}