{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OX7MJV5XQ7FOUFGDYTKNAE2CXW","short_pith_number":"pith:OX7MJV5X","schema_version":"1.0","canonical_sha256":"75fec4d7b787caea14c3c4d4d01342bda56175b5def09f7d85f11038f7dcb85c","source":{"kind":"arxiv","id":"2505.13026","version":3},"attestation_state":"computed","paper":{"title":"Step-wise Adaptive Integration of Supervised Fine-tuning and Reinforcement Learning for Task-Specific LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Erqu Qin, Fazhong Liu, Haojin Zhu, Harry Zheng, Jack Chen, Naruto Liu, Tian Dong, Xiao Wang, Yan Meng, Yuhan Luo","submitted_at":"2025-05-19T12:10:17Z","abstract_excerpt":"Large language models (LLMs) excel at mathematical reasoning and logical problem-solving. The current popular training paradigms primarily use supervised fine-tuning (SFT) and reinforcement learning (RL) to enhance the models' reasoning abilities. However, when using SFT or RL alone, there are respective challenges: SFT may suffer from overfitting, while RL is prone to mode collapse. The state-of-the-art methods have proposed hybrid training schemes. However, static switching faces challenges such as poor generalization across different tasks and high dependence on data quality. In response to"},"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":"2505.13026","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-19T12:10:17Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"006b4604d3bdfb86be8985f535abb3d95a6f01e0a91c3c202c7e31fe57553d56","abstract_canon_sha256":"6a79668bd3bde71c4d8374d8d88aca4e1b22effffb5d1236861d24ade599143b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:48:03.840128Z","signature_b64":"ef+Bv8uI1LoMo9O5wQgO/Mj1w1h3kOQU/MQtTmSnG+7XaFrwVNsleRJTzDvNqnKMZyVmyND2HQ9Kcebvlr9EAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"75fec4d7b787caea14c3c4d4d01342bda56175b5def09f7d85f11038f7dcb85c","last_reissued_at":"2026-07-05T11:48:03.839497Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:48:03.839497Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Step-wise Adaptive Integration of Supervised Fine-tuning and Reinforcement Learning for Task-Specific LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Erqu Qin, Fazhong Liu, Haojin Zhu, Harry Zheng, Jack Chen, Naruto Liu, Tian Dong, Xiao Wang, Yan Meng, Yuhan Luo","submitted_at":"2025-05-19T12:10:17Z","abstract_excerpt":"Large language models (LLMs) excel at mathematical reasoning and logical problem-solving. The current popular training paradigms primarily use supervised fine-tuning (SFT) and reinforcement learning (RL) to enhance the models' reasoning abilities. However, when using SFT or RL alone, there are respective challenges: SFT may suffer from overfitting, while RL is prone to mode collapse. The state-of-the-art methods have proposed hybrid training schemes. However, static switching faces challenges such as poor generalization across different tasks and high dependence on data quality. In response to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.13026","kind":"arxiv","version":3},"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/2505.13026/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":"2505.13026","created_at":"2026-07-05T11:48:03.839559+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.13026v3","created_at":"2026-07-05T11:48:03.839559+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.13026","created_at":"2026-07-05T11:48:03.839559+00:00"},{"alias_kind":"pith_short_12","alias_value":"OX7MJV5XQ7FO","created_at":"2026-07-05T11:48:03.839559+00:00"},{"alias_kind":"pith_short_16","alias_value":"OX7MJV5XQ7FOUFGD","created_at":"2026-07-05T11:48:03.839559+00:00"},{"alias_kind":"pith_short_8","alias_value":"OX7MJV5X","created_at":"2026-07-05T11:48:03.839559+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09932","citing_title":"When RL Fails after SFT: Rejuvenating Model Plasticity for Robust SFT-to-RL Handoff","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25571","citing_title":"AnE: Pushing the Reasoning Frontier of Multimodal LLMs via Anchor Evolution","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2507.01679","citing_title":"Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2508.00222","citing_title":"RL-PLUS: Countering Capability Boundary Collapse of LLMs in Reinforcement Learning with Hybrid-policy Optimization","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2508.07809","citing_title":"EvoCoT: Overcoming the Exploration Bottleneck in Reinforcement Learning","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15012","citing_title":"Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06387","citing_title":"Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23747","citing_title":"SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06387","citing_title":"Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06387","citing_title":"Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW","json":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW.json","graph_json":"https://pith.science/api/pith-number/OX7MJV5XQ7FOUFGDYTKNAE2CXW/graph.json","events_json":"https://pith.science/api/pith-number/OX7MJV5XQ7FOUFGDYTKNAE2CXW/events.json","paper":"https://pith.science/paper/OX7MJV5X"},"agent_actions":{"view_html":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW","download_json":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW.json","view_paper":"https://pith.science/paper/OX7MJV5X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.13026&json=true","fetch_graph":"https://pith.science/api/pith-number/OX7MJV5XQ7FOUFGDYTKNAE2CXW/graph.json","fetch_events":"https://pith.science/api/pith-number/OX7MJV5XQ7FOUFGDYTKNAE2CXW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW/action/storage_attestation","attest_author":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW/action/author_attestation","sign_citation":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW/action/citation_signature","submit_replication":"https://pith.science/pith/OX7MJV5XQ7FOUFGDYTKNAE2CXW/action/replication_record"}},"created_at":"2026-07-05T11:48:03.839559+00:00","updated_at":"2026-07-05T11:48:03.839559+00:00"}