{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AVQ6ZESPSIKOSRQITYM2Y7IGTR","short_pith_number":"pith:AVQ6ZESP","schema_version":"1.0","canonical_sha256":"0561ec924f9214e946089e19ac7d069c73f2a50d414d7270a15f9227912d8d4c","source":{"kind":"arxiv","id":"2503.02875","version":1},"attestation_state":"computed","paper":{"title":"The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Benyou Wang, Dong Yu, Haitao Mi, Jiahao Xu, Junying Chen, Ke Ji, Qiuzhi Liu, Tian Liang, Xiaoyuan Liu, Xingyu Chen, Zhaopeng Tu, Zhijie Wang, Zhiwei He","submitted_at":"2025-03-04T18:56:03Z","abstract_excerpt":"Improving the reasoning capabilities of large language models (LLMs) typically requires supervised fine-tuning with labeled data or computationally expensive sampling. We introduce Unsupervised Prefix Fine-Tuning (UPFT), which leverages the observation of Prefix Self-Consistency -- the shared initial reasoning steps across diverse solution trajectories -- to enhance LLM reasoning efficiency. By training exclusively on the initial prefix substrings (as few as 8 tokens), UPFT removes the need for labeled data or exhaustive sampling. Experiments on reasoning benchmarks show that UPFT matches the "},"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":"2503.02875","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-04T18:56:03Z","cross_cats_sorted":[],"title_canon_sha256":"baf657c970b5634c08f8d6bbf73ca7f8db4c585a6d96103d56f25f3ace100c5b","abstract_canon_sha256":"afc42a1a5b7d3b3607ea09e6b31c4f8956def63a2710ec2529928304b82aaee3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:24:16.092167Z","signature_b64":"3LhcGowCBw9kcpOxtO3o3Jf2UqR1rS5JTC5cDs+kTDzGmEevM1QeQlTNxLPE4Ucr/i8z/57898IItATD+0+PBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0561ec924f9214e946089e19ac7d069c73f2a50d414d7270a15f9227912d8d4c","last_reissued_at":"2026-07-05T10:24:16.091434Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:24:16.091434Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Benyou Wang, Dong Yu, Haitao Mi, Jiahao Xu, Junying Chen, Ke Ji, Qiuzhi Liu, Tian Liang, Xiaoyuan Liu, Xingyu Chen, Zhaopeng Tu, Zhijie Wang, Zhiwei He","submitted_at":"2025-03-04T18:56:03Z","abstract_excerpt":"Improving the reasoning capabilities of large language models (LLMs) typically requires supervised fine-tuning with labeled data or computationally expensive sampling. We introduce Unsupervised Prefix Fine-Tuning (UPFT), which leverages the observation of Prefix Self-Consistency -- the shared initial reasoning steps across diverse solution trajectories -- to enhance LLM reasoning efficiency. By training exclusively on the initial prefix substrings (as few as 8 tokens), UPFT removes the need for labeled data or exhaustive sampling. Experiments on reasoning benchmarks show that UPFT matches the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.02875","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/2503.02875/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":"2503.02875","created_at":"2026-07-05T10:24:16.091519+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.02875v1","created_at":"2026-07-05T10:24:16.091519+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.02875","created_at":"2026-07-05T10:24:16.091519+00:00"},{"alias_kind":"pith_short_12","alias_value":"AVQ6ZESPSIKO","created_at":"2026-07-05T10:24:16.091519+00:00"},{"alias_kind":"pith_short_16","alias_value":"AVQ6ZESPSIKOSRQI","created_at":"2026-07-05T10:24:16.091519+00:00"},{"alias_kind":"pith_short_8","alias_value":"AVQ6ZESP","created_at":"2026-07-05T10:24:16.091519+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2510.07962","citing_title":"LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11931","citing_title":"Learn to Think: Improving Multimodal Reasoning through Vision-Aware Self-Improvement Training","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR","json":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR.json","graph_json":"https://pith.science/api/pith-number/AVQ6ZESPSIKOSRQITYM2Y7IGTR/graph.json","events_json":"https://pith.science/api/pith-number/AVQ6ZESPSIKOSRQITYM2Y7IGTR/events.json","paper":"https://pith.science/paper/AVQ6ZESP"},"agent_actions":{"view_html":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR","download_json":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR.json","view_paper":"https://pith.science/paper/AVQ6ZESP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.02875&json=true","fetch_graph":"https://pith.science/api/pith-number/AVQ6ZESPSIKOSRQITYM2Y7IGTR/graph.json","fetch_events":"https://pith.science/api/pith-number/AVQ6ZESPSIKOSRQITYM2Y7IGTR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR/action/storage_attestation","attest_author":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR/action/author_attestation","sign_citation":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR/action/citation_signature","submit_replication":"https://pith.science/pith/AVQ6ZESPSIKOSRQITYM2Y7IGTR/action/replication_record"}},"created_at":"2026-07-05T10:24:16.091519+00:00","updated_at":"2026-07-05T10:24:16.091519+00:00"}