{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZKTL75HA3COLTOGNH67FLZQT2S","short_pith_number":"pith:ZKTL75HA","schema_version":"1.0","canonical_sha256":"caa6bff4e0d89cb9b8cd3fbe55e613d49823b4d7b3984813be47910754652248","source":{"kind":"arxiv","id":"2412.02595","version":2},"attestation_state":"computed","paper":{"title":"Nemotron-CC: Transforming Common Crawl into a Refined Long-Horizon Pretraining Dataset","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Brandon Norick, Bryan Catanzaro, Dan Su, Joseph Jennings, Kezhi Kong, Markus Kliegl, Mohammad Shoeybi, Mostofa Patwary, Ying Lin","submitted_at":"2024-12-03T17:28:50Z","abstract_excerpt":"Recent English Common Crawl datasets like FineWeb-Edu and DCLM achieved significant benchmark gains via aggressive model-based filtering, but at the cost of removing 90% of data. This limits their suitability for long token horizon training, such as 15T tokens for Llama 3.1. In this paper, we show how to achieve better trade-offs between accuracy and data quantity by a combination of classifier ensembling, synthetic data rephrasing, and reduced reliance on heuristic filters. When training 8B parameter models for 1T tokens, using a high-quality subset of our data improves MMLU by 5.6 over DCLM,"},"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":"2412.02595","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-03T17:28:50Z","cross_cats_sorted":[],"title_canon_sha256":"b93c4078a3f212e72c7381754cfade4304316cfa4050e61836babb3641c82c6f","abstract_canon_sha256":"ab1e42cfc62b8b9841448a5dae91ade59b4321e9500253c4427b6e5e278fd2f2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:16.347287Z","signature_b64":"XHZHolmO9ijZPVBrmV9VuknmNyzHUGDM0UGAcwL1Ntzc+Mswu35/zcrT8o6j48c6J1WoASbvrtdFDovJURSjCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"caa6bff4e0d89cb9b8cd3fbe55e613d49823b4d7b3984813be47910754652248","last_reissued_at":"2026-07-05T11:13:16.346764Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:16.346764Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Nemotron-CC: Transforming Common Crawl into a Refined Long-Horizon Pretraining Dataset","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Brandon Norick, Bryan Catanzaro, Dan Su, Joseph Jennings, Kezhi Kong, Markus Kliegl, Mohammad Shoeybi, Mostofa Patwary, Ying Lin","submitted_at":"2024-12-03T17:28:50Z","abstract_excerpt":"Recent English Common Crawl datasets like FineWeb-Edu and DCLM achieved significant benchmark gains via aggressive model-based filtering, but at the cost of removing 90% of data. This limits their suitability for long token horizon training, such as 15T tokens for Llama 3.1. In this paper, we show how to achieve better trade-offs between accuracy and data quantity by a combination of classifier ensembling, synthetic data rephrasing, and reduced reliance on heuristic filters. When training 8B parameter models for 1T tokens, using a high-quality subset of our data improves MMLU by 5.6 over DCLM,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.02595","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/2412.02595/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":"2412.02595","created_at":"2026-07-05T11:13:16.346828+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.02595v2","created_at":"2026-07-05T11:13:16.346828+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.02595","created_at":"2026-07-05T11:13:16.346828+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZKTL75HA3COL","created_at":"2026-07-05T11:13:16.346828+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZKTL75HA3COLTOGN","created_at":"2026-07-05T11:13:16.346828+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZKTL75HA","created_at":"2026-07-05T11:13:16.346828+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07778","citing_title":"Unlocking Latent Value: Taxonomy-Guided Recovery of High-Performing Data from Low-Tier Web Corpora","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31268","citing_title":"Mellum2 Technical Report","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2510.08008","citing_title":"Beyond Sunk Costs: Boosting LLM Pre-training Efficiency via Orthogonal Growth of Mixture-of-Experts","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19568","citing_title":"m3BERT: A Modern, Multi-lingual, Matryoshka Bidirectional Encoder","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2507.15640","citing_title":"Data Mixing Agent: Learning to Re-weight Domains for Continual Pre-training","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2510.26583","citing_title":"Emu3.5: Native Multimodal Models are World Learners","ref_index":84,"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":173,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25741","citing_title":"Scaling Latent Reasoning via Looped Language Models","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11405","citing_title":"20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03044","citing_title":"JoyAI-LLM Flash: Advancing Mid-Scale LLMs with Token Efficiency","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11405","citing_title":"20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10129","citing_title":"Synthetic Pre-Pre-Training Improves Language Model Robustness to Noisy Pre-Training Data","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2502.16982","citing_title":"Muon is Scalable for LLM Training","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2508.06471","citing_title":"GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05688","citing_title":"Attention Editing: A Versatile Framework for Cross-Architecture Attention Conversion","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S","json":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S.json","graph_json":"https://pith.science/api/pith-number/ZKTL75HA3COLTOGNH67FLZQT2S/graph.json","events_json":"https://pith.science/api/pith-number/ZKTL75HA3COLTOGNH67FLZQT2S/events.json","paper":"https://pith.science/paper/ZKTL75HA"},"agent_actions":{"view_html":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S","download_json":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S.json","view_paper":"https://pith.science/paper/ZKTL75HA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.02595&json=true","fetch_graph":"https://pith.science/api/pith-number/ZKTL75HA3COLTOGNH67FLZQT2S/graph.json","fetch_events":"https://pith.science/api/pith-number/ZKTL75HA3COLTOGNH67FLZQT2S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S/action/storage_attestation","attest_author":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S/action/author_attestation","sign_citation":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S/action/citation_signature","submit_replication":"https://pith.science/pith/ZKTL75HA3COLTOGNH67FLZQT2S/action/replication_record"}},"created_at":"2026-07-05T11:13:16.346828+00:00","updated_at":"2026-07-05T11:13:16.346828+00:00"}