{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:LPOQOCQMB6JSU6IXXVF2PZY6JA","short_pith_number":"pith:LPOQOCQM","schema_version":"1.0","canonical_sha256":"5bdd070a0c0f932a7917bd4ba7e71e480fc08e1258189fd3dc6ec86715fa0380","source":{"kind":"arxiv","id":"2009.14109","version":2},"attestation_state":"computed","paper":{"title":"Improving Low Compute Language Modeling with In-Domain Embedding Initialisation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Charles Welch, Jonathan K. Kummerfeld, Rada Mihalcea","submitted_at":"2020-09-29T15:48:58Z","abstract_excerpt":"Many NLP applications, such as biomedical data and technical support, have 10-100 million tokens of in-domain data and limited computational resources for learning from it. How should we train a language model in this scenario? Most language modeling research considers either a small dataset with a closed vocabulary (like the standard 1 million token Penn Treebank), or the whole web with byte-pair encoding. We show that for our target setting in English, initialising and freezing input embeddings using in-domain data can improve language model performance by providing a useful representation o"},"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":"2009.14109","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-09-29T15:48:58Z","cross_cats_sorted":[],"title_canon_sha256":"31db49cf7fd91f134aae5a5b4ea67dd19ce97a1b75a7fc227957bee8abb4898a","abstract_canon_sha256":"6049d9e02cb496dc993f023a1b8cc17f8c97bc0e1cb94118b483bda71db509ff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:39:17.247144Z","signature_b64":"Jf5Y415E4jwASh0HIdyUavjGzL9vYOWM7frfMUYEzNypTqEaOUnPxNq1IIpzkBEmXbAZaSPDhLBFy5ZeBhnwCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5bdd070a0c0f932a7917bd4ba7e71e480fc08e1258189fd3dc6ec86715fa0380","last_reissued_at":"2026-07-05T01:39:17.246754Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:39:17.246754Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving Low Compute Language Modeling with In-Domain Embedding Initialisation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Charles Welch, Jonathan K. Kummerfeld, Rada Mihalcea","submitted_at":"2020-09-29T15:48:58Z","abstract_excerpt":"Many NLP applications, such as biomedical data and technical support, have 10-100 million tokens of in-domain data and limited computational resources for learning from it. How should we train a language model in this scenario? Most language modeling research considers either a small dataset with a closed vocabulary (like the standard 1 million token Penn Treebank), or the whole web with byte-pair encoding. We show that for our target setting in English, initialising and freezing input embeddings using in-domain data can improve language model performance by providing a useful representation o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2009.14109","kind":"arxiv","version":2},"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/2009.14109/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":"2009.14109","created_at":"2026-07-05T01:39:17.246811+00:00"},{"alias_kind":"arxiv_version","alias_value":"2009.14109v2","created_at":"2026-07-05T01:39:17.246811+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2009.14109","created_at":"2026-07-05T01:39:17.246811+00:00"},{"alias_kind":"pith_short_12","alias_value":"LPOQOCQMB6JS","created_at":"2026-07-05T01:39:17.246811+00:00"},{"alias_kind":"pith_short_16","alias_value":"LPOQOCQMB6JSU6IX","created_at":"2026-07-05T01:39:17.246811+00:00"},{"alias_kind":"pith_short_8","alias_value":"LPOQOCQM","created_at":"2026-07-05T01:39:17.246811+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2412.13663","citing_title":"Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference","ref_index":197,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA","json":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA.json","graph_json":"https://pith.science/api/pith-number/LPOQOCQMB6JSU6IXXVF2PZY6JA/graph.json","events_json":"https://pith.science/api/pith-number/LPOQOCQMB6JSU6IXXVF2PZY6JA/events.json","paper":"https://pith.science/paper/LPOQOCQM"},"agent_actions":{"view_html":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA","download_json":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA.json","view_paper":"https://pith.science/paper/LPOQOCQM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2009.14109&json=true","fetch_graph":"https://pith.science/api/pith-number/LPOQOCQMB6JSU6IXXVF2PZY6JA/graph.json","fetch_events":"https://pith.science/api/pith-number/LPOQOCQMB6JSU6IXXVF2PZY6JA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA/action/storage_attestation","attest_author":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA/action/author_attestation","sign_citation":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA/action/citation_signature","submit_replication":"https://pith.science/pith/LPOQOCQMB6JSU6IXXVF2PZY6JA/action/replication_record"}},"created_at":"2026-07-05T01:39:17.246811+00:00","updated_at":"2026-07-05T01:39:17.246811+00:00"}