{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:QLELZKGR5MJIDMSP5A4MWKR6CR","short_pith_number":"pith:QLELZKGR","schema_version":"1.0","canonical_sha256":"82c8bca8d1eb1281b24fe838cb2a3e145410c8892e0c2f3ee79e8c954c3e28d0","source":{"kind":"arxiv","id":"2405.20541","version":1},"attestation_state":"computed","paper":{"title":"Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Cody Blakeney, Kartik Sreenivasan, Mansheej Paul, Matthew L. Leavitt, Max Marion, Zachary Ankner","submitted_at":"2024-05-30T23:50:20Z","abstract_excerpt":"In this work, we investigate whether small language models can determine high-quality subsets of large-scale text datasets that improve the performance of larger language models. While existing work has shown that pruning based on the perplexity of a larger model can yield high-quality data, we investigate whether smaller models can be used for perplexity-based pruning and how pruning is affected by the domain composition of the data being pruned. We demonstrate that for multiple dataset compositions, perplexity-based pruning of pretraining data can \\emph{significantly} improve downstream task"},"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":"2405.20541","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-30T23:50:20Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"ed94d694829432a9f9f9ba397cb3fe96ccc5fa75235806fe1ec854070ecd157c","abstract_canon_sha256":"9b18ef7b0aae1a98b4a753e804a781c2ae40f5a26c208b067e889cf0624d40f6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:25:38.776855Z","signature_b64":"ZxKzAoyY5ICrb/L9nOBS+Iqg0k5uBmRG0aNllzgxp3RmULqApE4/r0tz1Ko8MUtOJIkIE865vnpzhGQiPrbLAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"82c8bca8d1eb1281b24fe838cb2a3e145410c8892e0c2f3ee79e8c954c3e28d0","last_reissued_at":"2026-07-05T08:25:38.776459Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:25:38.776459Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Cody Blakeney, Kartik Sreenivasan, Mansheej Paul, Matthew L. Leavitt, Max Marion, Zachary Ankner","submitted_at":"2024-05-30T23:50:20Z","abstract_excerpt":"In this work, we investigate whether small language models can determine high-quality subsets of large-scale text datasets that improve the performance of larger language models. While existing work has shown that pruning based on the perplexity of a larger model can yield high-quality data, we investigate whether smaller models can be used for perplexity-based pruning and how pruning is affected by the domain composition of the data being pruned. We demonstrate that for multiple dataset compositions, perplexity-based pruning of pretraining data can \\emph{significantly} improve downstream task"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.20541","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/2405.20541/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":"2405.20541","created_at":"2026-07-05T08:25:38.776511+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.20541v1","created_at":"2026-07-05T08:25:38.776511+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.20541","created_at":"2026-07-05T08:25:38.776511+00:00"},{"alias_kind":"pith_short_12","alias_value":"QLELZKGR5MJI","created_at":"2026-07-05T08:25:38.776511+00:00"},{"alias_kind":"pith_short_16","alias_value":"QLELZKGR5MJIDMSP","created_at":"2026-07-05T08:25:38.776511+00:00"},{"alias_kind":"pith_short_8","alias_value":"QLELZKGR","created_at":"2026-07-05T08:25:38.776511+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28551","citing_title":"DataComp-VLM: Improved Open Datasets for Vision-Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28551","citing_title":"DataComp-VLM: Improved Open Datasets for Vision-Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28247","citing_title":"IRDS: Interpretable RLVR Data Selection via Verifier-Coupled Sparse Autoencoder Coverage","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2412.09413","citing_title":"Imitate, Explore, and Self-Improve: A Reproduction Report on Slow-thinking Reasoning Systems","ref_index":25,"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":12,"is_internal_anchor":false},{"citing_arxiv_id":"2603.18361","citing_title":"Synthetic Data Generation for Training Diversified Commonsense Reasoning Models","ref_index":1,"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":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11810","citing_title":"GRACE: A Dynamic Coreset Selection Framework for Large Language Model Optimization","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05227","citing_title":"Rethinking Data Curation in LLM Training: Online Reweighting Offers Better Generalization than Offline Methods","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR","json":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR.json","graph_json":"https://pith.science/api/pith-number/QLELZKGR5MJIDMSP5A4MWKR6CR/graph.json","events_json":"https://pith.science/api/pith-number/QLELZKGR5MJIDMSP5A4MWKR6CR/events.json","paper":"https://pith.science/paper/QLELZKGR"},"agent_actions":{"view_html":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR","download_json":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR.json","view_paper":"https://pith.science/paper/QLELZKGR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.20541&json=true","fetch_graph":"https://pith.science/api/pith-number/QLELZKGR5MJIDMSP5A4MWKR6CR/graph.json","fetch_events":"https://pith.science/api/pith-number/QLELZKGR5MJIDMSP5A4MWKR6CR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR/action/storage_attestation","attest_author":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR/action/author_attestation","sign_citation":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR/action/citation_signature","submit_replication":"https://pith.science/pith/QLELZKGR5MJIDMSP5A4MWKR6CR/action/replication_record"}},"created_at":"2026-07-05T08:25:38.776511+00:00","updated_at":"2026-07-05T08:25:38.776511+00:00"}