{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:2VSLHXZLA4T3HSYTCPVSDWTGDX","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f611460a07c52135d26e5d0aa86bf5d2c0167ea58dbd4c572cd6c471189765f1","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T09:17:51Z","title_canon_sha256":"d1f030a3df4a94c573b01b50ee1b517f6181a1e68243d22338561604cda508a0"},"schema_version":"1.0","source":{"id":"2605.13225","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13225","created_at":"2026-05-18T02:44:49Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13225v1","created_at":"2026-05-18T02:44:49Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13225","created_at":"2026-05-18T02:44:49Z"},{"alias_kind":"pith_short_12","alias_value":"2VSLHXZLA4T3","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"2VSLHXZLA4T3HSYT","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"2VSLHXZL","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:ab229bac2246cd60518ed8ad160d60bafe68ab9edee12af8c3e78b7568c97565","target":"graph","created_at":"2026-05-18T02:44:49Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"mixing yields larger improvements than hyperparameter tuning on both validation loss and downstream task accuracy, and the gap grows with model size. We quantify how much mixing helps: it boosts performance by an amount equivalent to 2--3× the unique target data on validation loss and 2--13× on downstream task accuracy, with the gain scaling steeply with model size."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the chosen mixing ratios are near-optimal and that English data supplies useful, non-conflicting signal for Arabic without introducing domain mismatch that would require separate controls."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Mixing auxiliary high-resource language data outperforms hyperparameter tuning in data-constrained bilingual pre-training, with gains equivalent to 2-13 times more unique target data."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Mixing high-resource language data outperforms hyperparameter tuning for low-resource pre-training."}],"snapshot_sha256":"e96f95120f13970c772a7c89914519d6f6af522e48cbbb2e517079cc756cfe12"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"10c417e86119014ae421d4f9069f9f5c0389799516d02066f38d0fb255324a0c"},"paper":{"abstract_excerpt":"For most languages of the world, language model pre-training operates in a data-constrained regime where models must repeat their training data many times, degrading generalization. Two remedies exist: aggressive hyperparameter tuning such as high weight decay, and mixing in data from a high-resource auxiliary language to directly aid the low-resource target. While hyperparameter tuning regularizes the model by shrinking weights to restrict network capacity, auxiliary data mixing uses a tunable mixing ratio to expand the training distribution and diversify the training signal with new knowledg","authors_text":"Anastasiia Sedova, Jes Frellsen, Louis B\\'ethune, Natalie Schluter, Paul Jeha, Pierre Ablin, Skyler Seto","cross_cats":[],"headline":"Mixing high-resource language data outperforms hyperparameter tuning for low-resource pre-training.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T09:17:51Z","title":"Mix, Don't Tune: Bilingual Pre-Training Outperforms Hyperparameter Search in Data-Constrained Settings"},"references":{"count":32,"internal_anchors":10,"resolved_work":32,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord","work_id":"b14fac55-a32e-45fb-b906-93ca894c02d8","year":null},{"cited_arxiv_id":"1911.02116","doi":"","is_internal_anchor":true,"ref_index":2,"title":"Unsupervised Cross-lingual Representation Learning at Scale","work_id":"32df83f5-69ea-418b-9a8d-03c2d6695b80","year":1911},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"arXiv preprint arXiv:2310.05492 , year=","work_id":"cc28e604-660c-4b03-853d-dec1e3dac25f","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"arXiv preprint arXiv:2403.08540 (2024)","work_id":"b3ccca34-2e12-48b4-ad09-521ec9797b0c","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt","work_id":"17ddc4a7-0ce2-4768-a1e4-b5d692e498e2","year":null}],"snapshot_sha256":"7ce333095a4a04d9987e66ed9e4c0e50751a278062957162ba4f7ccf4b0050c9"},"source":{"id":"2605.13225","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:13:37.436444Z","id":"a388c296-c723-4ebd-9e20-b9173d344f72","model_set":{"reader":"grok-4.3"},"one_line_summary":"Mixing auxiliary high-resource language data outperforms hyperparameter tuning in data-constrained bilingual pre-training, with gains equivalent to 2-13 times more unique target data.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Mixing high-resource language data outperforms hyperparameter tuning for low-resource pre-training.","strongest_claim":"mixing yields larger improvements than hyperparameter tuning on both validation loss and downstream task accuracy, and the gap grows with model size. We quantify how much mixing helps: it boosts performance by an amount equivalent to 2--3× the unique target data on validation loss and 2--13× on downstream task accuracy, with the gain scaling steeply with model size.","weakest_assumption":"That the chosen mixing ratios are near-optimal and that English data supplies useful, non-conflicting signal for Arabic without introducing domain mismatch that would require separate controls."}},"verdict_id":"a388c296-c723-4ebd-9e20-b9173d344f72"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:20866b46ec2bf86c84176bb67c892b2acff22bb03728df26b2aa9c6fb4bb26d4","target":"record","created_at":"2026-05-18T02:44:49Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f611460a07c52135d26e5d0aa86bf5d2c0167ea58dbd4c572cd6c471189765f1","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T09:17:51Z","title_canon_sha256":"d1f030a3df4a94c573b01b50ee1b517f6181a1e68243d22338561604cda508a0"},"schema_version":"1.0","source":{"id":"2605.13225","kind":"arxiv","version":1}},"canonical_sha256":"d564b3df2b0727b3cb1313eb21da661de40c6207e186cadfeb0f3b59d4385ca8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d564b3df2b0727b3cb1313eb21da661de40c6207e186cadfeb0f3b59d4385ca8","first_computed_at":"2026-05-18T02:44:49.635911Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:49.635911Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"kw1JmpwSp9mkJOgm5UOVmjMydBycpDTxsZNTDoS4kvUf9Y43BgJ8RGOL9ctv8D+tgx/9OAFaHjIPeZl4oqhFBw==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:49.636406Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13225","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:20866b46ec2bf86c84176bb67c892b2acff22bb03728df26b2aa9c6fb4bb26d4","sha256:ab229bac2246cd60518ed8ad160d60bafe68ab9edee12af8c3e78b7568c97565"],"state_sha256":"dcb6cae27a49d5ce959fef06dd23050a225f45cfab489329c4926bc46225d1aa"}