{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CB5VYPTNBLATOWOFLLGCAWB72N","short_pith_number":"pith:CB5VYPTN","schema_version":"1.0","canonical_sha256":"107b5c3e6d0ac13759c55acc20583fd36c841be29a97ed2810c33df6e769f129","source":{"kind":"arxiv","id":"2505.07686","version":2},"attestation_state":"computed","paper":{"title":"S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Chenxu Yang, Muzhi Dai, Qingyi Si","submitted_at":"2025-05-12T15:50:44Z","abstract_excerpt":"As Test-Time Scaling emerges as an active research focus in the large language model community, advanced post-training methods increasingly emphasize extending chain-of-thought (CoT) generation length, thereby enhancing reasoning capabilities to approach Deepseek R1-like reasoning models. However, recent studies reveal that reasoning models (even Qwen3) consistently exhibit excessive thought redundancy in CoT generation. This overthinking issue arises from the inherent limitations of conventional outcome-reward reinforcement learning, which systematically overlooks the regulation of intermedia"},"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":"2505.07686","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-05-12T15:50:44Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"3f5a187eb0fa036d92d430b8e2ddbe6ba5d3fd01a3437033341b094f490eb47b","abstract_canon_sha256":"e06991f835478ab215467373a201b5850f0e54c2f0eee7b53c31c0f7efab8e29"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:22.325849Z","signature_b64":"qsNErJlN3R99Z+EoJaroQiOqY71HWs8iPL9rtywNriIGml8Nprtp5QWrFD1JJowaQSGUd4TtIhcN3AFUwqkIDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"107b5c3e6d0ac13759c55acc20583fd36c841be29a97ed2810c33df6e769f129","last_reissued_at":"2026-07-05T11:04:22.325330Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:22.325330Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Chenxu Yang, Muzhi Dai, Qingyi Si","submitted_at":"2025-05-12T15:50:44Z","abstract_excerpt":"As Test-Time Scaling emerges as an active research focus in the large language model community, advanced post-training methods increasingly emphasize extending chain-of-thought (CoT) generation length, thereby enhancing reasoning capabilities to approach Deepseek R1-like reasoning models. However, recent studies reveal that reasoning models (even Qwen3) consistently exhibit excessive thought redundancy in CoT generation. This overthinking issue arises from the inherent limitations of conventional outcome-reward reinforcement learning, which systematically overlooks the regulation of intermedia"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.07686","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/2505.07686/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":"2505.07686","created_at":"2026-07-05T11:04:22.325403+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.07686v2","created_at":"2026-07-05T11:04:22.325403+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.07686","created_at":"2026-07-05T11:04:22.325403+00:00"},{"alias_kind":"pith_short_12","alias_value":"CB5VYPTNBLAT","created_at":"2026-07-05T11:04:22.325403+00:00"},{"alias_kind":"pith_short_16","alias_value":"CB5VYPTNBLATOWOF","created_at":"2026-07-05T11:04:22.325403+00:00"},{"alias_kind":"pith_short_8","alias_value":"CB5VYPTN","created_at":"2026-07-05T11:04:22.325403+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.07520","citing_title":"TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.05152","citing_title":"Reinforcement Learning from Rich Feedback with Distributional DAgger","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03503","citing_title":"ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03087","citing_title":"Learning to Solve, Forgetting to Retain: Correct-Set Turnover in RLVR","ref_index":99,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01249","citing_title":"Trust Region On-Policy Distillation","ref_index":126,"is_internal_anchor":false},{"citing_arxiv_id":"2509.26522","citing_title":"Entropy After </Think> for reasoning model early exiting","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":99,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11505","citing_title":"Selective Off-Policy Reference Tuning with Plan Guidance","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2503.16419","citing_title":"Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03128","citing_title":"Self-Distilled RLVR","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11505","citing_title":"Selective Off-Policy Reference Tuning with Plan Guidance","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27083","citing_title":"Co-Evolving Policy Distillation","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06165","citing_title":"Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost","ref_index":232,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07316","citing_title":"Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02913","citing_title":"Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16893","citing_title":"EasyVideoR1: Easier RL for Video Understanding","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20733","citing_title":"Near-Future Policy Optimization","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20659","citing_title":"GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N","json":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N.json","graph_json":"https://pith.science/api/pith-number/CB5VYPTNBLATOWOFLLGCAWB72N/graph.json","events_json":"https://pith.science/api/pith-number/CB5VYPTNBLATOWOFLLGCAWB72N/events.json","paper":"https://pith.science/paper/CB5VYPTN"},"agent_actions":{"view_html":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N","download_json":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N.json","view_paper":"https://pith.science/paper/CB5VYPTN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.07686&json=true","fetch_graph":"https://pith.science/api/pith-number/CB5VYPTNBLATOWOFLLGCAWB72N/graph.json","fetch_events":"https://pith.science/api/pith-number/CB5VYPTNBLATOWOFLLGCAWB72N/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N/action/storage_attestation","attest_author":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N/action/author_attestation","sign_citation":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N/action/citation_signature","submit_replication":"https://pith.science/pith/CB5VYPTNBLATOWOFLLGCAWB72N/action/replication_record"}},"created_at":"2026-07-05T11:04:22.325403+00:00","updated_at":"2026-07-05T11:04:22.325403+00:00"}