{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:2PXFC44XRAW7IAOFJBQFXJXBCI","short_pith_number":"pith:2PXFC44X","schema_version":"1.0","canonical_sha256":"d3ee517397882df401c548605ba6e11226570b717bf6081d07dffe6444c4b29a","source":{"kind":"arxiv","id":"2503.10460","version":4},"attestation_state":"computed","paper":{"title":"Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Fenrui Xiao, Haosheng Zou, Junchen Liu, Liang Wen, Lifu Tang, Qi An, Shousheng Jia, Xiangzheng Zhang, Xiaowei Lv, Xin He, Yimin Du, Yongchao Deng, Yunke Cai, Zhenyu Duan","submitted_at":"2025-03-13T15:29:22Z","abstract_excerpt":"This paper introduces Light-R1, an open-source suite for training long reasoning models using reproducible and cost-effective methodology. Given the proprietary nature of data used in the DeepSeek-R1 series, we develop an alternative approach leveraging exclusively public data and models. Our curriculum training progressively increases data difficulty, combined with multi-staged post-training. Our Light-R1-32B model, trained from Qwen2.5-32B-Instruct, outperforms DeepSeek-R1-Distill-Qwen-32B in math reasoning.\n  Experimental results show that this curriculum approach becomes more effective whe"},"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":"2503.10460","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-13T15:29:22Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"9dce75478fad8465aba69808edf9350e25539e7a6e4e48e35d19f17dcc661f17","abstract_canon_sha256":"7dfb6df145eb8aed49a534fa678228af46640ed618109efc3d2abcc8ba159621"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:54.251608Z","signature_b64":"ePStIAdUFgczbUFAzTarlueFRstY7jfLLtDJBYcnwofLTM0M6DIDYdkAOqo9W9MjqNP79WYlcxCAAbEDWBcjBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d3ee517397882df401c548605ba6e11226570b717bf6081d07dffe6444c4b29a","last_reissued_at":"2026-07-05T11:10:54.251056Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:54.251056Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Fenrui Xiao, Haosheng Zou, Junchen Liu, Liang Wen, Lifu Tang, Qi An, Shousheng Jia, Xiangzheng Zhang, Xiaowei Lv, Xin He, Yimin Du, Yongchao Deng, Yunke Cai, Zhenyu Duan","submitted_at":"2025-03-13T15:29:22Z","abstract_excerpt":"This paper introduces Light-R1, an open-source suite for training long reasoning models using reproducible and cost-effective methodology. Given the proprietary nature of data used in the DeepSeek-R1 series, we develop an alternative approach leveraging exclusively public data and models. Our curriculum training progressively increases data difficulty, combined with multi-staged post-training. Our Light-R1-32B model, trained from Qwen2.5-32B-Instruct, outperforms DeepSeek-R1-Distill-Qwen-32B in math reasoning.\n  Experimental results show that this curriculum approach becomes more effective whe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.10460","kind":"arxiv","version":4},"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/2503.10460/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":"2503.10460","created_at":"2026-07-05T11:10:54.251123+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.10460v4","created_at":"2026-07-05T11:10:54.251123+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.10460","created_at":"2026-07-05T11:10:54.251123+00:00"},{"alias_kind":"pith_short_12","alias_value":"2PXFC44XRAW7","created_at":"2026-07-05T11:10:54.251123+00:00"},{"alias_kind":"pith_short_16","alias_value":"2PXFC44XRAW7IAOF","created_at":"2026-07-05T11:10:54.251123+00:00"},{"alias_kind":"pith_short_8","alias_value":"2PXFC44X","created_at":"2026-07-05T11:10:54.251123+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":21,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02234","citing_title":"Purified OPSD: On-Policy Self-Distillation Without Losing How to Think","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11119","citing_title":"TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01168","citing_title":"Thinking Economically: A Hierarchical Framework for Adaptive-Complexity Reasoning in LLMs","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22567","citing_title":"LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2508.08636","citing_title":"InternBootcamp Technical Report: Boosting LLM Reasoning with Verifiable Task Scaling","ref_index":48,"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":21,"is_internal_anchor":false},{"citing_arxiv_id":"2602.01970","citing_title":"Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2509.25758","citing_title":"Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Training","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2510.03988","citing_title":"The Signal is in the Steps: Local Scoring for Reasoning Data Selection","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2510.04265","citing_title":"Don't Pass@k: A Bayesian Framework for Large Language Model Evaluation","ref_index":99,"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":21,"is_internal_anchor":false},{"citing_arxiv_id":"2505.22312","citing_title":"Skywork Open Reasoner 1 Technical Report","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2512.03847","citing_title":"DVPO: Distributional Value Modeling-based Policy Optimization for LLM Post-Training","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2504.14945","citing_title":"Learning to Reason under Off-Policy Guidance","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2504.20571","citing_title":"Reinforcement Learning for Reasoning in Large Language Models with One Training Example","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14054","citing_title":"Bad Seeing or Bad Thinking? Rewarding Perception for Multimodal Reasoning","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2503.16419","citing_title":"Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models","ref_index":191,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08441","citing_title":"DUET: Optimize Token-Budget Allocation for Reinforcement Learning with Verifiable Rewards","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08905","citing_title":"Forge: Quality-Aware Reinforcement Learning for NP-Hard Optimization in LLMs","ref_index":8,"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":159,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17614","citing_title":"Characterizing Model-Native Skills","ref_index":73,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI","json":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI.json","graph_json":"https://pith.science/api/pith-number/2PXFC44XRAW7IAOFJBQFXJXBCI/graph.json","events_json":"https://pith.science/api/pith-number/2PXFC44XRAW7IAOFJBQFXJXBCI/events.json","paper":"https://pith.science/paper/2PXFC44X"},"agent_actions":{"view_html":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI","download_json":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI.json","view_paper":"https://pith.science/paper/2PXFC44X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.10460&json=true","fetch_graph":"https://pith.science/api/pith-number/2PXFC44XRAW7IAOFJBQFXJXBCI/graph.json","fetch_events":"https://pith.science/api/pith-number/2PXFC44XRAW7IAOFJBQFXJXBCI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI/action/storage_attestation","attest_author":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI/action/author_attestation","sign_citation":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI/action/citation_signature","submit_replication":"https://pith.science/pith/2PXFC44XRAW7IAOFJBQFXJXBCI/action/replication_record"}},"created_at":"2026-07-05T11:10:54.251123+00:00","updated_at":"2026-07-05T11:10:54.251123+00:00"}