{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:X5JGHS4DWU7Y3MCEWDWHXQROQ2","short_pith_number":"pith:X5JGHS4D","schema_version":"1.0","canonical_sha256":"bf5263cb83b53f8db044b0ec7bc22e86b1e3d4b743500e8f5349b5c2eee365a4","source":{"kind":"arxiv","id":"2504.16677","version":1},"attestation_state":"computed","paper":{"title":"A Post-trainer's Guide to Multilingual Training Data: Uncovering Cross-lingual Transfer Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ahmet Ustun, Luisa Shimabucoro, Marzieh Fadaee, Sebastian Ruder","submitted_at":"2025-04-23T12:52:49Z","abstract_excerpt":"In order for large language models to be useful across the globe, they are fine-tuned to follow instructions on multilingual data. Despite the ubiquity of such post-training, a clear understanding of the dynamics that enable cross-lingual transfer remains elusive. This study examines cross-lingual transfer (CLT) dynamics in realistic post-training settings. We study two model families of up to 35B parameters in size trained on carefully controlled mixtures of multilingual data on three generative tasks with varying levels of complexity (summarization, instruction following, and mathematical re"},"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":"2504.16677","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-23T12:52:49Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9b49963368801337b66f04ed00230dec18f3611eea3e809c49bf14ccaae4d41c","abstract_canon_sha256":"79b04cbd8207362dc2f4b6256237b55098f3b1689172075565af629bc75d0f62"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:53:03.158207Z","signature_b64":"dHtAyVySBTR5QJkfm9hB6Ce8QSywBdsJv99JAOeQWPMmrX7ohiIk8g1QAPq8eD2UoFCnhuR3ddhoHTcmNTSlAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf5263cb83b53f8db044b0ec7bc22e86b1e3d4b743500e8f5349b5c2eee365a4","last_reissued_at":"2026-07-05T10:53:03.157723Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:53:03.157723Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Post-trainer's Guide to Multilingual Training Data: Uncovering Cross-lingual Transfer Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ahmet Ustun, Luisa Shimabucoro, Marzieh Fadaee, Sebastian Ruder","submitted_at":"2025-04-23T12:52:49Z","abstract_excerpt":"In order for large language models to be useful across the globe, they are fine-tuned to follow instructions on multilingual data. Despite the ubiquity of such post-training, a clear understanding of the dynamics that enable cross-lingual transfer remains elusive. This study examines cross-lingual transfer (CLT) dynamics in realistic post-training settings. We study two model families of up to 35B parameters in size trained on carefully controlled mixtures of multilingual data on three generative tasks with varying levels of complexity (summarization, instruction following, and mathematical re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.16677","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/2504.16677/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":"2504.16677","created_at":"2026-07-05T10:53:03.157785+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.16677v1","created_at":"2026-07-05T10:53:03.157785+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.16677","created_at":"2026-07-05T10:53:03.157785+00:00"},{"alias_kind":"pith_short_12","alias_value":"X5JGHS4DWU7Y","created_at":"2026-07-05T10:53:03.157785+00:00"},{"alias_kind":"pith_short_16","alias_value":"X5JGHS4DWU7Y3MCE","created_at":"2026-07-05T10:53:03.157785+00:00"},{"alias_kind":"pith_short_8","alias_value":"X5JGHS4D","created_at":"2026-07-05T10:53:03.157785+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.13286","citing_title":"English is Not All You Need: Systematically Exploring the Role of Multilinguality in LLM Post-Training","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20835","citing_title":"Parallel-SFT: Improving Zero-Shot Cross-Programming-Language Transfer for Code RL","ref_index":64,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2","json":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2.json","graph_json":"https://pith.science/api/pith-number/X5JGHS4DWU7Y3MCEWDWHXQROQ2/graph.json","events_json":"https://pith.science/api/pith-number/X5JGHS4DWU7Y3MCEWDWHXQROQ2/events.json","paper":"https://pith.science/paper/X5JGHS4D"},"agent_actions":{"view_html":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2","download_json":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2.json","view_paper":"https://pith.science/paper/X5JGHS4D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.16677&json=true","fetch_graph":"https://pith.science/api/pith-number/X5JGHS4DWU7Y3MCEWDWHXQROQ2/graph.json","fetch_events":"https://pith.science/api/pith-number/X5JGHS4DWU7Y3MCEWDWHXQROQ2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2/action/storage_attestation","attest_author":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2/action/author_attestation","sign_citation":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2/action/citation_signature","submit_replication":"https://pith.science/pith/X5JGHS4DWU7Y3MCEWDWHXQROQ2/action/replication_record"}},"created_at":"2026-07-05T10:53:03.157785+00:00","updated_at":"2026-07-05T10:53:03.157785+00:00"}