{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:WAUY3LLF2FZNXJRBJI4VNVOVSE","short_pith_number":"pith:WAUY3LLF","schema_version":"1.0","canonical_sha256":"b0298dad65d172dba6214a3956d5d5910fdd7a2338d757632d931bfe2abd7fca","source":{"kind":"arxiv","id":"2306.08756","version":1},"attestation_state":"computed","paper":{"title":"Recipes for Sequential Pre-training of Multilingual Encoder and Seq2Seq Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Andy Rosenbaum, Anna Rumshisky, Qin Lu, Saleh Soltan, Tobias Falke, Wael Hamza","submitted_at":"2023-06-14T21:41:52Z","abstract_excerpt":"Pre-trained encoder-only and sequence-to-sequence (seq2seq) models each have advantages, however training both model types from scratch is computationally expensive. We explore recipes to improve pre-training efficiency by initializing one model from the other. (1) Extracting the encoder from a seq2seq model, we show it under-performs a Masked Language Modeling (MLM) encoder, particularly on sequence labeling tasks. Variations of masking during seq2seq training, reducing the decoder size, and continuing with a small amount of MLM training do not close the gap. (2) Conversely, using an encoder "},"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":"2306.08756","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-06-14T21:41:52Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"622c28654d47ea2c902c88b07af106b5298514e8429545f5a6245c6a5fa6cf54","abstract_canon_sha256":"560154e712d53a69f024e7d69d15559e7f70a77d2969396f2e663c5bc868fc47"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:20:56.007010Z","signature_b64":"Gu+Si/1Y4j+hpjo08WbsHVELBNtFuQXqUUzFLDwJqJx6sgWZUl44AtNwakgBFyTyOtseCX28dpZQ6wLLftsuAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b0298dad65d172dba6214a3956d5d5910fdd7a2338d757632d931bfe2abd7fca","last_reissued_at":"2026-07-05T06:20:56.006480Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:20:56.006480Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Recipes for Sequential Pre-training of Multilingual Encoder and Seq2Seq Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Andy Rosenbaum, Anna Rumshisky, Qin Lu, Saleh Soltan, Tobias Falke, Wael Hamza","submitted_at":"2023-06-14T21:41:52Z","abstract_excerpt":"Pre-trained encoder-only and sequence-to-sequence (seq2seq) models each have advantages, however training both model types from scratch is computationally expensive. We explore recipes to improve pre-training efficiency by initializing one model from the other. (1) Extracting the encoder from a seq2seq model, we show it under-performs a Masked Language Modeling (MLM) encoder, particularly on sequence labeling tasks. Variations of masking during seq2seq training, reducing the decoder size, and continuing with a small amount of MLM training do not close the gap. (2) Conversely, using an encoder "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.08756","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/2306.08756/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":"2306.08756","created_at":"2026-07-05T06:20:56.006541+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.08756v1","created_at":"2026-07-05T06:20:56.006541+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.08756","created_at":"2026-07-05T06:20:56.006541+00:00"},{"alias_kind":"pith_short_12","alias_value":"WAUY3LLF2FZN","created_at":"2026-07-05T06:20:56.006541+00:00"},{"alias_kind":"pith_short_16","alias_value":"WAUY3LLF2FZNXJRB","created_at":"2026-07-05T06:20:56.006541+00:00"},{"alias_kind":"pith_short_8","alias_value":"WAUY3LLF","created_at":"2026-07-05T06:20:56.006541+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/WAUY3LLF2FZNXJRBJI4VNVOVSE","json":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE.json","graph_json":"https://pith.science/api/pith-number/WAUY3LLF2FZNXJRBJI4VNVOVSE/graph.json","events_json":"https://pith.science/api/pith-number/WAUY3LLF2FZNXJRBJI4VNVOVSE/events.json","paper":"https://pith.science/paper/WAUY3LLF"},"agent_actions":{"view_html":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE","download_json":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE.json","view_paper":"https://pith.science/paper/WAUY3LLF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.08756&json=true","fetch_graph":"https://pith.science/api/pith-number/WAUY3LLF2FZNXJRBJI4VNVOVSE/graph.json","fetch_events":"https://pith.science/api/pith-number/WAUY3LLF2FZNXJRBJI4VNVOVSE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE/action/storage_attestation","attest_author":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE/action/author_attestation","sign_citation":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE/action/citation_signature","submit_replication":"https://pith.science/pith/WAUY3LLF2FZNXJRBJI4VNVOVSE/action/replication_record"}},"created_at":"2026-07-05T06:20:56.006541+00:00","updated_at":"2026-07-05T06:20:56.006541+00:00"}