{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VSNBATHHLIHFBCSJ2XYVYUZIQO","short_pith_number":"pith:VSNBATHH","schema_version":"1.0","canonical_sha256":"ac9a104ce75a0e508a49d5f15c5328839d587cdbbcf88df2ea2e32a22b557cc7","source":{"kind":"arxiv","id":"2507.07562","version":1},"attestation_state":"computed","paper":{"title":"The Synergy Dilemma of Long-CoT SFT and RL: Investigating Post-Training Techniques for Reasoning VLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chaofan Tao, Fei Mi, Haoli Bai, Jiannan Wu, Jierun Chen, Kaican Li, Lei Zhu, Lewei Yao, Lifeng Shang, Lu Hou, Manyi Zhang, Qun Liu, Tiezheng Yu, Xiaohui Li","submitted_at":"2025-07-10T09:05:49Z","abstract_excerpt":"Large vision-language models (VLMs) increasingly adopt post-training techniques such as long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL) to elicit sophisticated reasoning. While these methods exhibit synergy in language-only models, their joint effectiveness in VLMs remains uncertain. We present a systematic investigation into the distinct roles and interplay of long-CoT SFT and RL across multiple multimodal reasoning benchmarks. We find that SFT improves performance on difficult questions by in-depth, structured reasoning, but introduces verbosity and d"},"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":"2507.07562","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-10T09:05:49Z","cross_cats_sorted":[],"title_canon_sha256":"91ed5dd52fa1e1455ffd5682dbc8fe32ec54dccb59547b52378104c09870feb9","abstract_canon_sha256":"044f1d6282f0dfa1ef4d8f4e6b4350d14ce367b4ca249c250505036dbf1fa90c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:00.858989Z","signature_b64":"zcH6kztHFMWEdZ35aYxL48o690XqwBQZPnTqg2r2q0lMmLqn4VLw5DQphy6qr+hgDHi1NayD/fWb/IPyUpt5Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac9a104ce75a0e508a49d5f15c5328839d587cdbbcf88df2ea2e32a22b557cc7","last_reissued_at":"2026-07-05T11:35:00.858468Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:00.858468Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Synergy Dilemma of Long-CoT SFT and RL: Investigating Post-Training Techniques for Reasoning VLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chaofan Tao, Fei Mi, Haoli Bai, Jiannan Wu, Jierun Chen, Kaican Li, Lei Zhu, Lewei Yao, Lifeng Shang, Lu Hou, Manyi Zhang, Qun Liu, Tiezheng Yu, Xiaohui Li","submitted_at":"2025-07-10T09:05:49Z","abstract_excerpt":"Large vision-language models (VLMs) increasingly adopt post-training techniques such as long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL) to elicit sophisticated reasoning. While these methods exhibit synergy in language-only models, their joint effectiveness in VLMs remains uncertain. We present a systematic investigation into the distinct roles and interplay of long-CoT SFT and RL across multiple multimodal reasoning benchmarks. We find that SFT improves performance on difficult questions by in-depth, structured reasoning, but introduces verbosity and d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.07562","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/2507.07562/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":"2507.07562","created_at":"2026-07-05T11:35:00.858529+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.07562v1","created_at":"2026-07-05T11:35:00.858529+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.07562","created_at":"2026-07-05T11:35:00.858529+00:00"},{"alias_kind":"pith_short_12","alias_value":"VSNBATHHLIHF","created_at":"2026-07-05T11:35:00.858529+00:00"},{"alias_kind":"pith_short_16","alias_value":"VSNBATHHLIHFBCSJ","created_at":"2026-07-05T11:35:00.858529+00:00"},{"alias_kind":"pith_short_8","alias_value":"VSNBATHH","created_at":"2026-07-05T11:35:00.858529+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25571","citing_title":"AnE: Pushing the Reasoning Frontier of Multimodal LLMs via Anchor Evolution","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2601.07389","citing_title":"On the Non-decoupling of Supervised Fine-tuning and Reinforcement Learning in Post-training","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2503.09567","citing_title":"Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models","ref_index":85,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO","json":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO.json","graph_json":"https://pith.science/api/pith-number/VSNBATHHLIHFBCSJ2XYVYUZIQO/graph.json","events_json":"https://pith.science/api/pith-number/VSNBATHHLIHFBCSJ2XYVYUZIQO/events.json","paper":"https://pith.science/paper/VSNBATHH"},"agent_actions":{"view_html":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO","download_json":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO.json","view_paper":"https://pith.science/paper/VSNBATHH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.07562&json=true","fetch_graph":"https://pith.science/api/pith-number/VSNBATHHLIHFBCSJ2XYVYUZIQO/graph.json","fetch_events":"https://pith.science/api/pith-number/VSNBATHHLIHFBCSJ2XYVYUZIQO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO/action/storage_attestation","attest_author":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO/action/author_attestation","sign_citation":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO/action/citation_signature","submit_replication":"https://pith.science/pith/VSNBATHHLIHFBCSJ2XYVYUZIQO/action/replication_record"}},"created_at":"2026-07-05T11:35:00.858529+00:00","updated_at":"2026-07-05T11:35:00.858529+00:00"}