{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Y4TP3YMSNROFUNPRF27EOR6MHZ","short_pith_number":"pith:Y4TP3YMS","schema_version":"1.0","canonical_sha256":"c726fde1926c5c5a35f12ebe4747cc3e6e2fea22ca9b743d916f586adbeeccc6","source":{"kind":"arxiv","id":"2606.01155","version":1},"attestation_state":"computed","paper":{"title":"When Data Is Scarce: Scaling Sparse Language Models with Repeated Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Boqian Wu, Decebal Constantin Mocanu, Elena Mocanu, Maurice van Keulen, Mykola Pechenizkiy, Patrik Okanovic, Qiao Xiao, Tomasz Sternal, Torsten Hoefler","submitted_at":"2026-05-31T10:51:18Z","abstract_excerpt":"Scaling laws for dense LLMs under infinite data are well explored, but how sparsity interacts with limited data is not. In this work, we study sparse training in data-constrained regimes where limited unique tokens require multi-epoch training. Our experiments span models up to 1.92B parameters in the fitting set, sparsity up to 93.75%, unique data budgets up to 2.6B tokens, and total training tokens up to 41.6B over 16 epochs; we further validate extrapolation on held-out dense-equivalent models up to 7.68B parameters. We find that: 1. Sparse scaling in data-limited settings: We introduce a s"},"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":"2606.01155","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-31T10:51:18Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2166b1f712da4277478ec011fe55edd8b099f042278a0aec6c9b96c99cb0ed6b","abstract_canon_sha256":"0d25c3aef4ffc3fed1eecc02835d9766d69dd8e35cfa8540fa01eaf1e9b0fef3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:25.253158Z","signature_b64":"BvYjldibs7BdoZcyaGtBnatfBkHmzirFHDKpCYZtOX6K5+x7A4PXhncbaxBGK48UOCvrsrVVxpdMNCBqFRhvAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c726fde1926c5c5a35f12ebe4747cc3e6e2fea22ca9b743d916f586adbeeccc6","last_reissued_at":"2026-06-02T02:04:25.252719Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:25.252719Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Data Is Scarce: Scaling Sparse Language Models with Repeated Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Boqian Wu, Decebal Constantin Mocanu, Elena Mocanu, Maurice van Keulen, Mykola Pechenizkiy, Patrik Okanovic, Qiao Xiao, Tomasz Sternal, Torsten Hoefler","submitted_at":"2026-05-31T10:51:18Z","abstract_excerpt":"Scaling laws for dense LLMs under infinite data are well explored, but how sparsity interacts with limited data is not. In this work, we study sparse training in data-constrained regimes where limited unique tokens require multi-epoch training. Our experiments span models up to 1.92B parameters in the fitting set, sparsity up to 93.75%, unique data budgets up to 2.6B tokens, and total training tokens up to 41.6B over 16 epochs; we further validate extrapolation on held-out dense-equivalent models up to 7.68B parameters. We find that: 1. Sparse scaling in data-limited settings: We introduce a s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.01155","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/2606.01155/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":"2606.01155","created_at":"2026-06-02T02:04:25.252777+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.01155v1","created_at":"2026-06-02T02:04:25.252777+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.01155","created_at":"2026-06-02T02:04:25.252777+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y4TP3YMSNROF","created_at":"2026-06-02T02:04:25.252777+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y4TP3YMSNROFUNPR","created_at":"2026-06-02T02:04:25.252777+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y4TP3YMS","created_at":"2026-06-02T02:04:25.252777+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ","json":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ.json","graph_json":"https://pith.science/api/pith-number/Y4TP3YMSNROFUNPRF27EOR6MHZ/graph.json","events_json":"https://pith.science/api/pith-number/Y4TP3YMSNROFUNPRF27EOR6MHZ/events.json","paper":"https://pith.science/paper/Y4TP3YMS"},"agent_actions":{"view_html":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ","download_json":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ.json","view_paper":"https://pith.science/paper/Y4TP3YMS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.01155&json=true","fetch_graph":"https://pith.science/api/pith-number/Y4TP3YMSNROFUNPRF27EOR6MHZ/graph.json","fetch_events":"https://pith.science/api/pith-number/Y4TP3YMSNROFUNPRF27EOR6MHZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ/action/storage_attestation","attest_author":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ/action/author_attestation","sign_citation":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ/action/citation_signature","submit_replication":"https://pith.science/pith/Y4TP3YMSNROFUNPRF27EOR6MHZ/action/replication_record"}},"created_at":"2026-06-02T02:04:25.252777+00:00","updated_at":"2026-06-02T02:04:25.252777+00:00"}