{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:R3TB4SIMRLJABRCQMNKGP4N2WE","short_pith_number":"pith:R3TB4SIM","schema_version":"1.0","canonical_sha256":"8ee61e490c8ad200c450635467f1bab1246039f11d1333d9a7f0be0d7a8e4f45","source":{"kind":"arxiv","id":"2506.20920","version":1},"attestation_state":"computed","paper":{"title":"FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir Hossein Kargaran, Bettina Messmer, Colin Raffel, Guilherme Penedo, Hynek Kydl\\'i\\v{c}ek, Leandro Von Werra, Martin Jaggi, Negar Foroutan, Thomas Wolf, Vinko Sabol\\v{c}ec","submitted_at":"2025-06-26T01:01:47Z","abstract_excerpt":"Pre-training state-of-the-art large language models (LLMs) requires vast amounts of clean and diverse text data. While the open development of large high-quality English pre-training datasets has seen substantial recent progress, training performant multilingual LLMs remains a challenge, in large part due to the inherent difficulty of tailoring filtering and deduplication pipelines to a large number of languages. In this work, we introduce a new pre-training dataset curation pipeline based on FineWeb that can be automatically adapted to support any language. We extensively ablate our pipeline "},"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":"2506.20920","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-26T01:01:47Z","cross_cats_sorted":[],"title_canon_sha256":"cdbbcfa019bb781512ddd45a3f61f6656425ef4583643150bb35767be6a605ea","abstract_canon_sha256":"91b46ea12280b413a971ce18b93fb1552a398ede374e43c40d06b9f8ba31cc95"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:36.161857Z","signature_b64":"6FV8E4d5XKGMEHyAJUVkUbe5SjGyyyMhUx6gfuQO5yrW5KWkVrU5LikNrf2J/cLXUhDOAKV8JCb+VhVWucaHAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8ee61e490c8ad200c450635467f1bab1246039f11d1333d9a7f0be0d7a8e4f45","last_reissued_at":"2026-07-05T11:27:36.161374Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:36.161374Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir Hossein Kargaran, Bettina Messmer, Colin Raffel, Guilherme Penedo, Hynek Kydl\\'i\\v{c}ek, Leandro Von Werra, Martin Jaggi, Negar Foroutan, Thomas Wolf, Vinko Sabol\\v{c}ec","submitted_at":"2025-06-26T01:01:47Z","abstract_excerpt":"Pre-training state-of-the-art large language models (LLMs) requires vast amounts of clean and diverse text data. While the open development of large high-quality English pre-training datasets has seen substantial recent progress, training performant multilingual LLMs remains a challenge, in large part due to the inherent difficulty of tailoring filtering and deduplication pipelines to a large number of languages. In this work, we introduce a new pre-training dataset curation pipeline based on FineWeb that can be automatically adapted to support any language. We extensively ablate our pipeline "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.20920","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/2506.20920/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":"2506.20920","created_at":"2026-07-05T11:27:36.161435+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.20920v1","created_at":"2026-07-05T11:27:36.161435+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.20920","created_at":"2026-07-05T11:27:36.161435+00:00"},{"alias_kind":"pith_short_12","alias_value":"R3TB4SIMRLJA","created_at":"2026-07-05T11:27:36.161435+00:00"},{"alias_kind":"pith_short_16","alias_value":"R3TB4SIMRLJABRCQ","created_at":"2026-07-05T11:27:36.161435+00:00"},{"alias_kind":"pith_short_8","alias_value":"R3TB4SIM","created_at":"2026-07-05T11:27:36.161435+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25821","citing_title":"SARA: Unlocking Multilingual Knowledge in Mixture-of-Experts via Semantically Anchored Routing Alignment","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.23566","citing_title":"LangMAP: A Language-Adaptive Approach to Tokenization","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22722","citing_title":"moBERTo: A Modern Encoder for Portuguese via Continued Pretraining of ModernBERT","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20945","citing_title":"Grouped Query Experts: Mixture-of-Experts on GQA Self-Attention","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08545","citing_title":"Ishigaki-IDS: An Open-Weight Verifier-Aware Model for Information Delivery Specification Drafting in Building Information Modeling","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25263","citing_title":"Mimir: Large-scale Multilingual Concept Modeling","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28057","citing_title":"MultiHashFormer: Hash-based Generative Language Models","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23885","citing_title":"Multilingual Knowledge Transfer under Data Constraints via Lexical Interventions","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18083","citing_title":"A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$\\Delta$ Integration into Upcycled MoE","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18232","citing_title":"SomaliWeb v1: A Quality-Filtered Somali Web Corpus with a Matched Tokenizer and a Public Language-Identification Benchmark","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12715","citing_title":"Scaling Laws for Mixture Pretraining Under Data Constraints","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12715","citing_title":"Scaling Laws for Mixture Pretraining Under Data Constraints","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13225","citing_title":"Mix, Don't Tune: Bilingual Pre-Training Outperforms Hyperparameter Search in Data-Constrained Settings","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13521","citing_title":"Granite Embedding Multilingual R2 Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2508.06471","citing_title":"GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE","json":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE.json","graph_json":"https://pith.science/api/pith-number/R3TB4SIMRLJABRCQMNKGP4N2WE/graph.json","events_json":"https://pith.science/api/pith-number/R3TB4SIMRLJABRCQMNKGP4N2WE/events.json","paper":"https://pith.science/paper/R3TB4SIM"},"agent_actions":{"view_html":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE","download_json":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE.json","view_paper":"https://pith.science/paper/R3TB4SIM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.20920&json=true","fetch_graph":"https://pith.science/api/pith-number/R3TB4SIMRLJABRCQMNKGP4N2WE/graph.json","fetch_events":"https://pith.science/api/pith-number/R3TB4SIMRLJABRCQMNKGP4N2WE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE/action/storage_attestation","attest_author":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE/action/author_attestation","sign_citation":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE/action/citation_signature","submit_replication":"https://pith.science/pith/R3TB4SIMRLJABRCQMNKGP4N2WE/action/replication_record"}},"created_at":"2026-07-05T11:27:36.161435+00:00","updated_at":"2026-07-05T11:27:36.161435+00:00"}