{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:FNJJPLKLFP3O676R4FC4LVTPNA","short_pith_number":"pith:FNJJPLKL","schema_version":"1.0","canonical_sha256":"2b5297ad4b2bf6ef7fd1e145c5d66f6832763243db391199463d0d8f8def052d","source":{"kind":"arxiv","id":"2402.09668","version":1},"attestation_state":"computed","paper":{"title":"How to Train Data-Efficient LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Benjamin Coleman, Derek Zhiyuan Cheng, Ed H. Chi, James Caverlee, Jianmo Ni, Julian McAuley, Lichan Hong, Noveen Sachdeva, Wang-Cheng Kang","submitted_at":"2024-02-15T02:27:57Z","abstract_excerpt":"The training of large language models (LLMs) is expensive. In this paper, we study data-efficient approaches for pre-training LLMs, i.e., techniques that aim to optimize the Pareto frontier of model quality and training resource/data consumption. We seek to understand the tradeoffs associated with data selection routines based on (i) expensive-to-compute data-quality estimates, and (ii) maximization of coverage and diversity-based measures in the feature space. Our first technique, Ask-LLM, leverages the zero-shot reasoning capabilities of instruction-tuned LLMs to directly assess the quality "},"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":"2402.09668","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-15T02:27:57Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"ae9abe1ec66c1688fd78f24ac60cf1858ee87e2c502d5ac3fee71f8b3584eeff","abstract_canon_sha256":"964442b9127ef1c6081b96f70ab22367480921a1e1b3b1b509d9982f52c8d197"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:45:34.619387Z","signature_b64":"g1ppw1yOv3cbZlfGsH+31Vtr5GtqvpZxOIxHMCtyZjE7CKsfrsmJ7mLkvwkmUyru0mhlBR7raZBXShI6LSH/AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2b5297ad4b2bf6ef7fd1e145c5d66f6832763243db391199463d0d8f8def052d","last_reissued_at":"2026-07-05T07:45:34.618944Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:45:34.618944Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How to Train Data-Efficient LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Benjamin Coleman, Derek Zhiyuan Cheng, Ed H. Chi, James Caverlee, Jianmo Ni, Julian McAuley, Lichan Hong, Noveen Sachdeva, Wang-Cheng Kang","submitted_at":"2024-02-15T02:27:57Z","abstract_excerpt":"The training of large language models (LLMs) is expensive. In this paper, we study data-efficient approaches for pre-training LLMs, i.e., techniques that aim to optimize the Pareto frontier of model quality and training resource/data consumption. We seek to understand the tradeoffs associated with data selection routines based on (i) expensive-to-compute data-quality estimates, and (ii) maximization of coverage and diversity-based measures in the feature space. Our first technique, Ask-LLM, leverages the zero-shot reasoning capabilities of instruction-tuned LLMs to directly assess the quality "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.09668","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/2402.09668/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":"2402.09668","created_at":"2026-07-05T07:45:34.619005+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.09668v1","created_at":"2026-07-05T07:45:34.619005+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.09668","created_at":"2026-07-05T07:45:34.619005+00:00"},{"alias_kind":"pith_short_12","alias_value":"FNJJPLKLFP3O","created_at":"2026-07-05T07:45:34.619005+00:00"},{"alias_kind":"pith_short_16","alias_value":"FNJJPLKLFP3O676R","created_at":"2026-07-05T07:45:34.619005+00:00"},{"alias_kind":"pith_short_8","alias_value":"FNJJPLKL","created_at":"2026-07-05T07:45:34.619005+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.30337","citing_title":"Efficient Test-Time Finetuning of LLMs via Convex Reconstruction and Gradient Caching","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07597","citing_title":"Repetition Mismatch: Why Data Mixture Experiments Don't Scale and How to Fix Them","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2503.19786","citing_title":"Gemma 3 Technical Report","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2406.11794","citing_title":"DataComp-LM: In search of the next generation of training sets for language models","ref_index":157,"is_internal_anchor":false},{"citing_arxiv_id":"2403.17297","citing_title":"InternLM2 Technical Report","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06230","citing_title":"Safactory: A Scalable Agentic Infrastructure for Training Trustworthy Autonomous Intelligence","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12397","citing_title":"KoCo: Conditioning Language Model Pre-training on Knowledge Coordinates","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08519","citing_title":"Cram Less to Fit More: Training Data Pruning Improves Memorization of Facts","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07769","citing_title":"An Empirical Study on Influence-Based Pretraining Data Selection for Code Large Language Models","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06900","citing_title":"Accelerated Relax-and-Round for Concave Coverage Problems","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06230","citing_title":"Safactory: A Scalable Agentic Infrastructure for Training Trustworthy Autonomous Intelligence","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06901","citing_title":"Reflections and New Directions for Human-Centered Large Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2303.18223","citing_title":"A Survey of Large Language Models","ref_index":237,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16979","citing_title":"DOSE: Data Selection for Multi-Modal LLMs via Off-the-Shelf Models","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA","json":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA.json","graph_json":"https://pith.science/api/pith-number/FNJJPLKLFP3O676R4FC4LVTPNA/graph.json","events_json":"https://pith.science/api/pith-number/FNJJPLKLFP3O676R4FC4LVTPNA/events.json","paper":"https://pith.science/paper/FNJJPLKL"},"agent_actions":{"view_html":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA","download_json":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA.json","view_paper":"https://pith.science/paper/FNJJPLKL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.09668&json=true","fetch_graph":"https://pith.science/api/pith-number/FNJJPLKLFP3O676R4FC4LVTPNA/graph.json","fetch_events":"https://pith.science/api/pith-number/FNJJPLKLFP3O676R4FC4LVTPNA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA/action/storage_attestation","attest_author":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA/action/author_attestation","sign_citation":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA/action/citation_signature","submit_replication":"https://pith.science/pith/FNJJPLKLFP3O676R4FC4LVTPNA/action/replication_record"}},"created_at":"2026-07-05T07:45:34.619005+00:00","updated_at":"2026-07-05T07:45:34.619005+00:00"}