{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LOIXRSKSUXJNJRKR2AGJJTVWGV","short_pith_number":"pith:LOIXRSKS","schema_version":"1.0","canonical_sha256":"5b9178c952a5d2d4c551d00c94ceb635498eac0179b4a637e46092addf27a607","source":{"kind":"arxiv","id":"2501.11747","version":2},"attestation_state":"computed","paper":{"title":"Optimizing Pretraining Data Mixtures with LLM-Estimated Utility","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bhargavi Paranjape, Frank Zhang, Mike Lewis, Punit Singh Koura, Todor Mihaylov, William Held","submitted_at":"2025-01-20T21:10:22Z","abstract_excerpt":"Large Language Models improve with increasing amounts of high-quality training data. However, leveraging larger datasets requires balancing quality, quantity, and diversity across sources. After evaluating nine baseline methods under both compute- and data-constrained scenarios, we find token-count heuristics outperform manual and learned mixes, indicating that simple approaches accounting for dataset size and diversity are surprisingly effective. Building on this insight, we propose two complementary approaches: UtiliMax, which extends token-based heuristics by incorporating utility estimates"},"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":"2501.11747","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-01-20T21:10:22Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d07b3388a046460d4b0ff2c9f1dc458ef1bf18e61868e43603e6fbdba489dbb9","abstract_canon_sha256":"2fe64aeae756bf581327e7fe1de00e0b2264dbe74bab397f231b0d3a7ed0e658"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:04:42.014811Z","signature_b64":"sN/dR/BAAgsjimzRpbs5R4xixyW+1XxyFW5KHdVt6Vk7O//zwwSKLQY0SWfzTg0YLXOBAjV7E7ctyplYawtRCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5b9178c952a5d2d4c551d00c94ceb635498eac0179b4a637e46092addf27a607","last_reissued_at":"2026-07-05T10:04:42.014341Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:04:42.014341Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimizing Pretraining Data Mixtures with LLM-Estimated Utility","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bhargavi Paranjape, Frank Zhang, Mike Lewis, Punit Singh Koura, Todor Mihaylov, William Held","submitted_at":"2025-01-20T21:10:22Z","abstract_excerpt":"Large Language Models improve with increasing amounts of high-quality training data. However, leveraging larger datasets requires balancing quality, quantity, and diversity across sources. After evaluating nine baseline methods under both compute- and data-constrained scenarios, we find token-count heuristics outperform manual and learned mixes, indicating that simple approaches accounting for dataset size and diversity are surprisingly effective. Building on this insight, we propose two complementary approaches: UtiliMax, which extends token-based heuristics by incorporating utility estimates"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.11747","kind":"arxiv","version":2},"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/2501.11747/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":"2501.11747","created_at":"2026-07-05T10:04:42.014400+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.11747v2","created_at":"2026-07-05T10:04:42.014400+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.11747","created_at":"2026-07-05T10:04:42.014400+00:00"},{"alias_kind":"pith_short_12","alias_value":"LOIXRSKSUXJN","created_at":"2026-07-05T10:04:42.014400+00:00"},{"alias_kind":"pith_short_16","alias_value":"LOIXRSKSUXJNJRKR","created_at":"2026-07-05T10:04:42.014400+00:00"},{"alias_kind":"pith_short_8","alias_value":"LOIXRSKS","created_at":"2026-07-05T10:04:42.014400+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.16380","citing_title":"Data Mixing for Large Language Models Pretraining: A Survey and Outlook","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV","json":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV.json","graph_json":"https://pith.science/api/pith-number/LOIXRSKSUXJNJRKR2AGJJTVWGV/graph.json","events_json":"https://pith.science/api/pith-number/LOIXRSKSUXJNJRKR2AGJJTVWGV/events.json","paper":"https://pith.science/paper/LOIXRSKS"},"agent_actions":{"view_html":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV","download_json":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV.json","view_paper":"https://pith.science/paper/LOIXRSKS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.11747&json=true","fetch_graph":"https://pith.science/api/pith-number/LOIXRSKSUXJNJRKR2AGJJTVWGV/graph.json","fetch_events":"https://pith.science/api/pith-number/LOIXRSKSUXJNJRKR2AGJJTVWGV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV/action/storage_attestation","attest_author":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV/action/author_attestation","sign_citation":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV/action/citation_signature","submit_replication":"https://pith.science/pith/LOIXRSKSUXJNJRKR2AGJJTVWGV/action/replication_record"}},"created_at":"2026-07-05T10:04:42.014400+00:00","updated_at":"2026-07-05T10:04:42.014400+00:00"}