{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TTWJ3XTA4UJANH7CXAHT5FNIYO","short_pith_number":"pith:TTWJ3XTA","schema_version":"1.0","canonical_sha256":"9cec9dde60e512069fe2b80f3e95a8c3a82dc7784ec333f3533c5f14ed1b2957","source":{"kind":"arxiv","id":"2503.07572","version":1},"attestation_state":"computed","paper":{"title":"Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Amrith Setlur, Aviral Kumar, Edward Emanuel Beeching, Lewis Tunstall, Matthew Y. R. Yang, Ruslan Salakhutdinov, Yuxiao Qu","submitted_at":"2025-03-10T17:40:43Z","abstract_excerpt":"Training models to effectively use test-time compute is crucial for improving the reasoning performance of LLMs. Current methods mostly do so via fine-tuning on search traces or running RL with 0/1 outcome reward, but do these approaches efficiently utilize test-time compute? Would these approaches continue to scale as the budget improves? In this paper, we try to answer these questions. We formalize the problem of optimizing test-time compute as a meta-reinforcement learning (RL) problem, which provides a principled perspective on spending test-time compute. This perspective enables us to vie"},"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":"2503.07572","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-10T17:40:43Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"ba21c14a13242817fd97b1d13d9e755bb0bb5cbe71627ec5b825eb564d60b7a0","abstract_canon_sha256":"c596a5a41de87c46792daa7150fb271a5dba4714fdf8c016fd0eaecc7e53a997"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:28:07.324493Z","signature_b64":"C6a6BbB8aK7IgHUTBN2SGfSbuqYwo2H/O2WKpw+9Udr0ZdYIB+8ToKOGOLBTIq0MK6L0nCjOPq3v0ymP+LULDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9cec9dde60e512069fe2b80f3e95a8c3a82dc7784ec333f3533c5f14ed1b2957","last_reissued_at":"2026-07-05T10:28:07.323637Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:28:07.323637Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Amrith Setlur, Aviral Kumar, Edward Emanuel Beeching, Lewis Tunstall, Matthew Y. R. Yang, Ruslan Salakhutdinov, Yuxiao Qu","submitted_at":"2025-03-10T17:40:43Z","abstract_excerpt":"Training models to effectively use test-time compute is crucial for improving the reasoning performance of LLMs. Current methods mostly do so via fine-tuning on search traces or running RL with 0/1 outcome reward, but do these approaches efficiently utilize test-time compute? Would these approaches continue to scale as the budget improves? In this paper, we try to answer these questions. We formalize the problem of optimizing test-time compute as a meta-reinforcement learning (RL) problem, which provides a principled perspective on spending test-time compute. This perspective enables us to vie"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.07572","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/2503.07572/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":"2503.07572","created_at":"2026-07-05T10:28:07.323754+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.07572v1","created_at":"2026-07-05T10:28:07.323754+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.07572","created_at":"2026-07-05T10:28:07.323754+00:00"},{"alias_kind":"pith_short_12","alias_value":"TTWJ3XTA4UJA","created_at":"2026-07-05T10:28:07.323754+00:00"},{"alias_kind":"pith_short_16","alias_value":"TTWJ3XTA4UJANH7C","created_at":"2026-07-05T10:28:07.323754+00:00"},{"alias_kind":"pith_short_8","alias_value":"TTWJ3XTA","created_at":"2026-07-05T10:28:07.323754+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23640","citing_title":"Learning Process Rewards via Success Visitation Matching for Efficient RL","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":157,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31748","citing_title":"Addressing Over-Refusal in LLMs with Competing Rewards","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08665","citing_title":"Hint Tuning: Less Data Makes Better Reasoners","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24396","citing_title":"Understanding and Mitigating Premature Confidence for Better LLM Reasoning","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2509.13332","citing_title":"Explicit Reasoning Makes Better Judges: A Systematic Study on Accuracy, Efficiency, and Robustness","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2602.06475","citing_title":"Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoning","ref_index":9,"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":33,"is_internal_anchor":false},{"citing_arxiv_id":"2503.16419","citing_title":"Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models","ref_index":146,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08665","citing_title":"Hint Tuning: Less Data Makes Better Reasoners","ref_index":20,"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":157,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02290","citing_title":"Distilling Long-CoT Reasoning through Collaborative Step-wise Multi-Teacher Decoding","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07851","citing_title":"ReRec: Reasoning-Augmented LLM-based Recommendation Assistant via Reinforcement Fine-tuning","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18839","citing_title":"One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15830","citing_title":"Placing Puzzle Pieces Where They Matter: A Question Augmentation Framework for Reinforcement Learning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20659","citing_title":"GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO","json":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO.json","graph_json":"https://pith.science/api/pith-number/TTWJ3XTA4UJANH7CXAHT5FNIYO/graph.json","events_json":"https://pith.science/api/pith-number/TTWJ3XTA4UJANH7CXAHT5FNIYO/events.json","paper":"https://pith.science/paper/TTWJ3XTA"},"agent_actions":{"view_html":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO","download_json":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO.json","view_paper":"https://pith.science/paper/TTWJ3XTA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.07572&json=true","fetch_graph":"https://pith.science/api/pith-number/TTWJ3XTA4UJANH7CXAHT5FNIYO/graph.json","fetch_events":"https://pith.science/api/pith-number/TTWJ3XTA4UJANH7CXAHT5FNIYO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO/action/storage_attestation","attest_author":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO/action/author_attestation","sign_citation":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO/action/citation_signature","submit_replication":"https://pith.science/pith/TTWJ3XTA4UJANH7CXAHT5FNIYO/action/replication_record"}},"created_at":"2026-07-05T10:28:07.323754+00:00","updated_at":"2026-07-05T10:28:07.323754+00:00"}