{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:B5MLVYKFFLRIURBVZDHABBHPO6","short_pith_number":"pith:B5MLVYKF","schema_version":"1.0","canonical_sha256":"0f58bae1452ae28a4435c8ce0084ef77922f07c81a4ccbddfb48d5080516988a","source":{"kind":"arxiv","id":"2506.13284","version":1},"attestation_state":"computed","paper":{"title":"AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergy","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bryan Catanzaro, Chankyu Lee, Mohammad Shoeybi, Wei Ping, Yang Chen, Zhuolin Yang, Zihan Liu","submitted_at":"2025-06-16T09:27:48Z","abstract_excerpt":"In this work, we investigate the synergy between supervised fine-tuning (SFT) and reinforcement learning (RL) in developing strong reasoning models. We begin by curating the SFT training data through two scaling strategies: increasing the number of collected prompts and the number of generated responses per prompt. Both approaches yield notable improvements in reasoning performance, with scaling the number of prompts resulting in more substantial gains. We then explore the following questions regarding the synergy between SFT and RL: (i) Does a stronger SFT model consistently lead to better fi"},"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":"2506.13284","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-16T09:27:48Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"06e4a19a61155bc10483adb86555a831fae823abfeef854ed850e8f73b62eb00","abstract_canon_sha256":"8289938f64885930eb98e36819a3677bd950fafda64496a2c714625c562e1028"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:22:08.063301Z","signature_b64":"CS/r2R2b7mIDVeW/ah18Bm3XkPFHpjjyTKHwWpX8zs1RJllTAOEuN1Xt3+pvyvAEV6PD8O4kJ5V+krizcBLGAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0f58bae1452ae28a4435c8ce0084ef77922f07c81a4ccbddfb48d5080516988a","last_reissued_at":"2026-07-05T11:22:08.062822Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:22:08.062822Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergy","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bryan Catanzaro, Chankyu Lee, Mohammad Shoeybi, Wei Ping, Yang Chen, Zhuolin Yang, Zihan Liu","submitted_at":"2025-06-16T09:27:48Z","abstract_excerpt":"In this work, we investigate the synergy between supervised fine-tuning (SFT) and reinforcement learning (RL) in developing strong reasoning models. We begin by curating the SFT training data through two scaling strategies: increasing the number of collected prompts and the number of generated responses per prompt. Both approaches yield notable improvements in reasoning performance, with scaling the number of prompts resulting in more substantial gains. We then explore the following questions regarding the synergy between SFT and RL: (i) Does a stronger SFT model consistently lead to better fi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.13284","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/2506.13284/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":"2506.13284","created_at":"2026-07-05T11:22:08.062881+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.13284v1","created_at":"2026-07-05T11:22:08.062881+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.13284","created_at":"2026-07-05T11:22:08.062881+00:00"},{"alias_kind":"pith_short_12","alias_value":"B5MLVYKFFLRI","created_at":"2026-07-05T11:22:08.062881+00:00"},{"alias_kind":"pith_short_16","alias_value":"B5MLVYKFFLRIURBV","created_at":"2026-07-05T11:22:08.062881+00:00"},{"alias_kind":"pith_short_8","alias_value":"B5MLVYKF","created_at":"2026-07-05T11:22:08.062881+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":17,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11634","citing_title":"Architecture-Aware Reinforcement Learning Makes Sliding-Window Attention Competitive in Math Reasoning","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01249","citing_title":"Trust Region On-Policy Distillation","ref_index":255,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18643","citing_title":"Post-Trained MoE Can Skip Half Experts via Self-Distillation","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29303","citing_title":"Entropy-KL Divergence-based Token Masking: A Novel Approach for Selective Fine-tuning of Large Language Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18600","citing_title":"MapTab: Are MLLMs Ready for Multi-Criteria Route Planning in Heterogeneous Graphs?","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07721","citing_title":"Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18643","citing_title":"Post-Trained MoE Can Skip Half Experts via Self-Distillation","ref_index":17,"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":13,"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":100,"is_internal_anchor":false},{"citing_arxiv_id":"2512.20856","citing_title":"NVIDIA Nemotron 3: Efficient and Open Intelligence","ref_index":170,"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":9,"is_internal_anchor":false},{"citing_arxiv_id":"2602.01203","citing_title":"Attention Sink Forges Native MoE in Attention Layers: Sink-Aware Training to Address Head Collapse","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18600","citing_title":"MapTab: Are MLLMs Ready for Multi-Criteria Route Planning in Heterogeneous Graphs?","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25741","citing_title":"Scaling Latent Reasoning via Looped Language Models","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14164","citing_title":"How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize Student-Consistent SFT Data","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07721","citing_title":"Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17614","citing_title":"Characterizing Model-Native Skills","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6","json":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6.json","graph_json":"https://pith.science/api/pith-number/B5MLVYKFFLRIURBVZDHABBHPO6/graph.json","events_json":"https://pith.science/api/pith-number/B5MLVYKFFLRIURBVZDHABBHPO6/events.json","paper":"https://pith.science/paper/B5MLVYKF"},"agent_actions":{"view_html":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6","download_json":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6.json","view_paper":"https://pith.science/paper/B5MLVYKF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.13284&json=true","fetch_graph":"https://pith.science/api/pith-number/B5MLVYKFFLRIURBVZDHABBHPO6/graph.json","fetch_events":"https://pith.science/api/pith-number/B5MLVYKFFLRIURBVZDHABBHPO6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6/action/storage_attestation","attest_author":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6/action/author_attestation","sign_citation":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6/action/citation_signature","submit_replication":"https://pith.science/pith/B5MLVYKFFLRIURBVZDHABBHPO6/action/replication_record"}},"created_at":"2026-07-05T11:22:08.062881+00:00","updated_at":"2026-07-05T11:22:08.062881+00:00"}