{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:KHQV23NUBYIMYY6AK2KOPWUPRW","short_pith_number":"pith:KHQV23NU","schema_version":"1.0","canonical_sha256":"51e15d6db40e10cc63c05694e7da8f8d8ed7379aad99bb3e1a5a6e983bd7350d","source":{"kind":"arxiv","id":"2502.07374","version":2},"attestation_state":"computed","paper":{"title":"LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Dacheng Li, Eric Tang, Ion Stoica, Joseph E. Gonzalez, Kourosh Hakhamaneshi, Matei Zaharia, Shishir G. Patil, Shiyi Cao, Shu Liu, Sumanth Hegde, Tyler Griggs, Xiangxi Mo","submitted_at":"2025-02-11T08:48:48Z","abstract_excerpt":"Large reasoning models (LRMs) tackle complex reasoning problems by following long chain-of-thoughts (Long CoT) that incorporate reflection, backtracking, and self-validation. However, the training techniques and data requirements to elicit Long CoT remain poorly understood. In this work, we find that a Large Language model (LLM) can effectively learn Long CoT reasoning through data-efficient supervised fine-tuning (SFT) and parameter-efficient low-rank adaptation (LoRA). With just 17k long CoT training samples, the Qwen2.5-32B-Instruct model achieves significant improvements on a wide range of"},"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":"2502.07374","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-11T08:48:48Z","cross_cats_sorted":[],"title_canon_sha256":"38433bc43b1964b25753751921159d6f88067c9deb8d782c920d72510900c70d","abstract_canon_sha256":"7acc7a20d8d7d8f03c4cf67cd97076aa19e75768b488f24a4878adb9843c14b9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:15:49.043845Z","signature_b64":"uAuGzzBOTgOhgnJf8jYKUeYsRZ7QfigJGD5sBGqgXhLLV/DoNXWz5WohFLtjezhr4c58HQ8eqUC7jMWbZ5A6Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"51e15d6db40e10cc63c05694e7da8f8d8ed7379aad99bb3e1a5a6e983bd7350d","last_reissued_at":"2026-07-05T10:15:49.043285Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:15:49.043285Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Dacheng Li, Eric Tang, Ion Stoica, Joseph E. Gonzalez, Kourosh Hakhamaneshi, Matei Zaharia, Shishir G. Patil, Shiyi Cao, Shu Liu, Sumanth Hegde, Tyler Griggs, Xiangxi Mo","submitted_at":"2025-02-11T08:48:48Z","abstract_excerpt":"Large reasoning models (LRMs) tackle complex reasoning problems by following long chain-of-thoughts (Long CoT) that incorporate reflection, backtracking, and self-validation. However, the training techniques and data requirements to elicit Long CoT remain poorly understood. In this work, we find that a Large Language model (LLM) can effectively learn Long CoT reasoning through data-efficient supervised fine-tuning (SFT) and parameter-efficient low-rank adaptation (LoRA). With just 17k long CoT training samples, the Qwen2.5-32B-Instruct model achieves significant improvements on a wide range of"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.07374","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/2502.07374/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":"2502.07374","created_at":"2026-07-05T10:15:49.043388+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.07374v2","created_at":"2026-07-05T10:15:49.043388+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.07374","created_at":"2026-07-05T10:15:49.043388+00:00"},{"alias_kind":"pith_short_12","alias_value":"KHQV23NUBYIM","created_at":"2026-07-05T10:15:49.043388+00:00"},{"alias_kind":"pith_short_16","alias_value":"KHQV23NUBYIMYY6A","created_at":"2026-07-05T10:15:49.043388+00:00"},{"alias_kind":"pith_short_8","alias_value":"KHQV23NU","created_at":"2026-07-05T10:15:49.043388+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":19,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23404","citing_title":"ReasoningLens: Hierarchical Visualization and Diagnostic Auditing for Large Reasoning Models","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06188","citing_title":"The Tell-Tale Norm: $\\ell_2$ Magnitude as a Signal for Reasoning Dynamics in Large Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01084","citing_title":"Can Agents Generalize to the Open World? Unveiling the Fragility of Static Training in Tool Use","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31492","citing_title":"LinTree: Improving LLM Reasoning with Explicitly Structured Search Histories","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00206","citing_title":"Quantized Reasoning Models Think They Need to Think Longer, but They Do Not","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11470","citing_title":"The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes","ref_index":134,"is_internal_anchor":false},{"citing_arxiv_id":"2504.01990","citing_title":"Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems","ref_index":156,"is_internal_anchor":false},{"citing_arxiv_id":"2507.15778","citing_title":"Stabilizing Knowledge, Promoting Reasoning: Dual-Token Constraints for RLVR","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18814","citing_title":"A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2506.18841","citing_title":"LongWriter-Zero: Mastering Ultra-Long Text Generation via Reinforcement Learning","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18814","citing_title":"A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2504.01943","citing_title":"OpenCodeReasoning: Advancing Data Distillation for Competitive Coding","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14249","citing_title":"Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2506.06941","citing_title":"The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2506.01939","citing_title":"Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09678","citing_title":"Absurd World: A Simple Yet Powerful Method to Absurdify the Real-world for Probing LLM Reasoning Capabilities","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06882","citing_title":"How Well Do LLMs Perform on the Simplest Long-Chain Reasoning Tasks: An Empirical Study on the Equivalence Class Problem","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17433","citing_title":"Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning","ref_index":81,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW","json":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW.json","graph_json":"https://pith.science/api/pith-number/KHQV23NUBYIMYY6AK2KOPWUPRW/graph.json","events_json":"https://pith.science/api/pith-number/KHQV23NUBYIMYY6AK2KOPWUPRW/events.json","paper":"https://pith.science/paper/KHQV23NU"},"agent_actions":{"view_html":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW","download_json":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW.json","view_paper":"https://pith.science/paper/KHQV23NU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.07374&json=true","fetch_graph":"https://pith.science/api/pith-number/KHQV23NUBYIMYY6AK2KOPWUPRW/graph.json","fetch_events":"https://pith.science/api/pith-number/KHQV23NUBYIMYY6AK2KOPWUPRW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW/action/storage_attestation","attest_author":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW/action/author_attestation","sign_citation":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW/action/citation_signature","submit_replication":"https://pith.science/pith/KHQV23NUBYIMYY6AK2KOPWUPRW/action/replication_record"}},"created_at":"2026-07-05T10:15:49.043388+00:00","updated_at":"2026-07-05T10:15:49.043388+00:00"}