{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EDPRMAQQNICDYZZWOC7ULZDMTQ","short_pith_number":"pith:EDPRMAQQ","schema_version":"1.0","canonical_sha256":"20df1602106a043c673670bf45e46c9c28ac9c128c239f55e486d5dd96db5230","source":{"kind":"arxiv","id":"2507.01679","version":3},"attestation_state":"computed","paper":{"title":"Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Edoardo M. Ponti, Ivan Titov, Tianhao Cheng, Yinghui Xu, Zeyu Huang, Zihan Qiu, Zili Wang","submitted_at":"2025-07-02T13:04:09Z","abstract_excerpt":"Existing LLMs-post-training techniques are broadly categorized into supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT). Each paradigm presents a distinct trade-off: (1) SFT excels at mimicking demonstration data, but can lead to problematic generalization as a form of behavior cloning. (2) Conversely, RFT can significantly enhance a model's performance but is prone to learning unexpected behaviors, and its performance is sensitive to the initial policy. In this paper, we propose a unified view of these methods and introduce Prefix-RFT, a hybrid approach that synergizes learning f"},"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":"2507.01679","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-02T13:04:09Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a52ab5ac2b944d0ba470efe078c004a67e7c87c49ef7dbabe131d33a6db54783","abstract_canon_sha256":"8390f738c802f7ba62a5207ad61243bb666cad5e61ba66c8b82c99959d80b745"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:21.918166Z","signature_b64":"TtZjEXy2uVzNUb+RUuPP3HhQDkR0ph4F9RVsDbbx434IIN1BE4hAf3nGfci6QD654KTBnuoc3KjRT1ariMT2Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"20df1602106a043c673670bf45e46c9c28ac9c128c239f55e486d5dd96db5230","last_reissued_at":"2026-05-20T00:00:21.917416Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:21.917416Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Edoardo M. Ponti, Ivan Titov, Tianhao Cheng, Yinghui Xu, Zeyu Huang, Zihan Qiu, Zili Wang","submitted_at":"2025-07-02T13:04:09Z","abstract_excerpt":"Existing LLMs-post-training techniques are broadly categorized into supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT). Each paradigm presents a distinct trade-off: (1) SFT excels at mimicking demonstration data, but can lead to problematic generalization as a form of behavior cloning. (2) Conversely, RFT can significantly enhance a model's performance but is prone to learning unexpected behaviors, and its performance is sensitive to the initial policy. In this paper, we propose a unified view of these methods and introduce Prefix-RFT, a hybrid approach that synergizes learning f"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.01679","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/2507.01679/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":"2507.01679","created_at":"2026-05-20T00:00:21.917534+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.01679v3","created_at":"2026-05-20T00:00:21.917534+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.01679","created_at":"2026-05-20T00:00:21.917534+00:00"},{"alias_kind":"pith_short_12","alias_value":"EDPRMAQQNICD","created_at":"2026-05-20T00:00:21.917534+00:00"},{"alias_kind":"pith_short_16","alias_value":"EDPRMAQQNICDYZZW","created_at":"2026-05-20T00:00:21.917534+00:00"},{"alias_kind":"pith_short_8","alias_value":"EDPRMAQQ","created_at":"2026-05-20T00:00:21.917534+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":12,"sample":[{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":216,"is_internal_anchor":true},{"citing_arxiv_id":"2603.11321","citing_title":"Hindsight-Anchored Policy Optimization: Turning Failure into Feedback in Sparse Reward Settings","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15012","citing_title":"Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06387","citing_title":"Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08666","citing_title":"The Cancellation Hypothesis in Critic-Free RL: From Outcome Rewards to Token Credits","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2604.23747","citing_title":"SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06387","citing_title":"Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13010","citing_title":"Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13010","citing_title":"Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06387","citing_title":"Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2604.16893","citing_title":"EasyVideoR1: Easier RL for Video Understanding","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2604.20733","citing_title":"Near-Future Policy Optimization","ref_index":9,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ","json":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ.json","graph_json":"https://pith.science/api/pith-number/EDPRMAQQNICDYZZWOC7ULZDMTQ/graph.json","events_json":"https://pith.science/api/pith-number/EDPRMAQQNICDYZZWOC7ULZDMTQ/events.json","paper":"https://pith.science/paper/EDPRMAQQ"},"agent_actions":{"view_html":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ","download_json":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ.json","view_paper":"https://pith.science/paper/EDPRMAQQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.01679&json=true","fetch_graph":"https://pith.science/api/pith-number/EDPRMAQQNICDYZZWOC7ULZDMTQ/graph.json","fetch_events":"https://pith.science/api/pith-number/EDPRMAQQNICDYZZWOC7ULZDMTQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ/action/storage_attestation","attest_author":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ/action/author_attestation","sign_citation":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ/action/citation_signature","submit_replication":"https://pith.science/pith/EDPRMAQQNICDYZZWOC7ULZDMTQ/action/replication_record"}},"created_at":"2026-05-20T00:00:21.917534+00:00","updated_at":"2026-05-20T00:00:21.917534+00:00"}