{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PFJRCOLJWJWAQBBQWIMFNT2BUU","short_pith_number":"pith:PFJRCOLJ","schema_version":"1.0","canonical_sha256":"7953113969b26c080430b21856cf41a50746807e9112c1cf66314b00b657da94","source":{"kind":"arxiv","id":"2605.28042","version":1},"attestation_state":"computed","paper":{"title":"Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Liu O. Martin, Lucas Bandarkar, Nanyun Peng","submitted_at":"2026-05-27T06:46:42Z","abstract_excerpt":"Modern large language models (LLMs) achieve state-of-the-art machine translation performance, but they do so as broad generalists largely trained for many tasks and capabilities unrelated to translation. Thus, they are heavily overparameterized for this task, resulting in excessive memory and compute requirements. In this paper, we present a method for aggressively pruning experts from modern mixture-of-experts LLMs while incurring negligible degradation in translation quality. Our approach exploits expert specialization and the separability of multilingual capabilities in LLMs to identify exp"},"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":"2605.28042","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-27T06:46:42Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"964df77af2f26ffbabeb2615520ad91309d726df14b381e3e80b04e8d585af78","abstract_canon_sha256":"ab5552df341a5fe2e45ef22c61926a3b772d6707aeb029269874087d4d9c2147"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:56.863381Z","signature_b64":"9JrTuwipKyKhnbnxEBvGMYRBzUEcBPn4gazpukgUJFFxpUSzjjFESmGeU8jSaM/jgpfwNzE6+YrzDRtOZcQ9DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7953113969b26c080430b21856cf41a50746807e9112c1cf66314b00b657da94","last_reissued_at":"2026-05-28T01:04:56.862829Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:56.862829Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Liu O. Martin, Lucas Bandarkar, Nanyun Peng","submitted_at":"2026-05-27T06:46:42Z","abstract_excerpt":"Modern large language models (LLMs) achieve state-of-the-art machine translation performance, but they do so as broad generalists largely trained for many tasks and capabilities unrelated to translation. Thus, they are heavily overparameterized for this task, resulting in excessive memory and compute requirements. In this paper, we present a method for aggressively pruning experts from modern mixture-of-experts LLMs while incurring negligible degradation in translation quality. Our approach exploits expert specialization and the separability of multilingual capabilities in LLMs to identify exp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28042","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/2605.28042/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":"2605.28042","created_at":"2026-05-28T01:04:56.862913+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28042v1","created_at":"2026-05-28T01:04:56.862913+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28042","created_at":"2026-05-28T01:04:56.862913+00:00"},{"alias_kind":"pith_short_12","alias_value":"PFJRCOLJWJWA","created_at":"2026-05-28T01:04:56.862913+00:00"},{"alias_kind":"pith_short_16","alias_value":"PFJRCOLJWJWAQBBQ","created_at":"2026-05-28T01:04:56.862913+00:00"},{"alias_kind":"pith_short_8","alias_value":"PFJRCOLJ","created_at":"2026-05-28T01:04:56.862913+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/PFJRCOLJWJWAQBBQWIMFNT2BUU","json":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU.json","graph_json":"https://pith.science/api/pith-number/PFJRCOLJWJWAQBBQWIMFNT2BUU/graph.json","events_json":"https://pith.science/api/pith-number/PFJRCOLJWJWAQBBQWIMFNT2BUU/events.json","paper":"https://pith.science/paper/PFJRCOLJ"},"agent_actions":{"view_html":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU","download_json":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU.json","view_paper":"https://pith.science/paper/PFJRCOLJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28042&json=true","fetch_graph":"https://pith.science/api/pith-number/PFJRCOLJWJWAQBBQWIMFNT2BUU/graph.json","fetch_events":"https://pith.science/api/pith-number/PFJRCOLJWJWAQBBQWIMFNT2BUU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU/action/storage_attestation","attest_author":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU/action/author_attestation","sign_citation":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU/action/citation_signature","submit_replication":"https://pith.science/pith/PFJRCOLJWJWAQBBQWIMFNT2BUU/action/replication_record"}},"created_at":"2026-05-28T01:04:56.862913+00:00","updated_at":"2026-05-28T01:04:56.862913+00:00"}