{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:42FNTVNGRGPDPGCLDZRLBGDJOZ","short_pith_number":"pith:42FNTVNG","schema_version":"1.0","canonical_sha256":"e68ad9d5a6899e37984b1e62b098697672f8bfa48de329b4fb0bf6ca0b714960","source":{"kind":"arxiv","id":"2207.14255","version":1},"attestation_state":"computed","paper":{"title":"Efficient Training of Language Models to Fill in the Middle","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Christine McLeavey, Heewoo Jun, Jerry Tworek, John Schulman, Mark Chen, Mohammad Bavarian, Nikolas Tezak","submitted_at":"2022-07-28T17:40:47Z","abstract_excerpt":"We show that autoregressive language models can learn to infill text after we apply a straightforward transformation to the dataset, which simply moves a span of text from the middle of a document to its end. While this data augmentation has garnered much interest in recent years, we provide extensive evidence that training models with a large fraction of data transformed in this way does not harm the original left-to-right generative capability, as measured by perplexity and sampling evaluations across a wide range of scales. Given the usefulness, simplicity, and efficiency of training models"},"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":"2207.14255","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-07-28T17:40:47Z","cross_cats_sorted":[],"title_canon_sha256":"e3f17dec2398ee2bed0424c1881217774b214a1303ffa79fbf895246d41b0e20","abstract_canon_sha256":"5c730c8b906cbb6fa41ef63a66f39d68103c6ce78288ef7e6e78d63b08517e89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T00:35:37.704631Z","signature_b64":"UooRznMuMzlmbRmNk3htWg0hBm1b4o5py8Gb+xffuP/IkwGxj9Gh+/U1Px0jD/L6qf6SjCvIhg78XKpVzIEuCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e68ad9d5a6899e37984b1e62b098697672f8bfa48de329b4fb0bf6ca0b714960","last_reissued_at":"2026-05-18T00:35:37.704002Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T00:35:37.704002Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Efficient Training of Language Models to Fill in the Middle","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Christine McLeavey, Heewoo Jun, Jerry Tworek, John Schulman, Mark Chen, Mohammad Bavarian, Nikolas Tezak","submitted_at":"2022-07-28T17:40:47Z","abstract_excerpt":"We show that autoregressive language models can learn to infill text after we apply a straightforward transformation to the dataset, which simply moves a span of text from the middle of a document to its end. While this data augmentation has garnered much interest in recent years, we provide extensive evidence that training models with a large fraction of data transformed in this way does not harm the original left-to-right generative capability, as measured by perplexity and sampling evaluations across a wide range of scales. Given the usefulness, simplicity, and efficiency of training models"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2207.14255","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":""},"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":"2207.14255","created_at":"2026-05-18T00:35:37.704085+00:00"},{"alias_kind":"arxiv_version","alias_value":"2207.14255v1","created_at":"2026-05-18T00:35:37.704085+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2207.14255","created_at":"2026-05-18T00:35:37.704085+00:00"},{"alias_kind":"pith_short_12","alias_value":"42FNTVNGRGPD","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_16","alias_value":"42FNTVNGRGPDPGCL","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_8","alias_value":"42FNTVNG","created_at":"2026-05-18T12:33:33.725879+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":21,"internal_anchor_count":21,"sample":[{"citing_arxiv_id":"2510.03843","citing_title":"Smart Paste: Automatically Fixing Copy/Paste for Google Developers","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2510.20129","citing_title":"SAID: Safety-Aware Intent Defense via Prefix Probing for Large Language Models","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2211.15089","citing_title":"Continuous diffusion for categorical data","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2511.07752","citing_title":"Back to the Future: The Role of Past and Future Context Predictability in Incremental Language Production","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2506.17298","citing_title":"Mercury: Ultra-Fast Language Models Based on Diffusion","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2311.16867","citing_title":"The Falcon Series of Open Language Models","ref_index":247,"is_internal_anchor":true},{"citing_arxiv_id":"2602.06759","citing_title":"\"Tab, Tab, Bug\": Security Pitfalls of Next Edit Suggestions in AI-Integrated IDEs","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2204.05999","citing_title":"InCoder: A Generative Model for Code Infilling and Synthesis","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2501.15383","citing_title":"Qwen2.5-1M Technical Report","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2501.15383","citing_title":"Qwen2.5-1M Technical Report","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2306.11644","citing_title":"Textbooks Are All You Need","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2402.19173","citing_title":"StarCoder 2 and The Stack v2: The Next Generation","ref_index":160,"is_internal_anchor":true},{"citing_arxiv_id":"2604.27780","citing_title":"RuC: HDL-Agnostic Rule Completion Benchmark Generation","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08744","citing_title":"MeshFIM: Local Low-Poly Mesh Editing via Fill-in-the-Middle Autoregressive Generation","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05902","citing_title":"Evaluating Non-English Developer Support in Machine Learning for Software Engineering","ref_index":110,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04894","citing_title":"SynConfRoute: Syntax-Aware Routing for Efficient Code Completion with Small CodeLLMs","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2502.09992","citing_title":"Large Language Diffusion Models","ref_index":122,"is_internal_anchor":true},{"citing_arxiv_id":"2305.06161","citing_title":"StarCoder: may the source be with you!","ref_index":148,"is_internal_anchor":true},{"citing_arxiv_id":"2604.04380","citing_title":"CPT: Controllable and Editable Design Variations with Language Models","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2401.14196","citing_title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2409.12186","citing_title":"Qwen2.5-Coder Technical Report","ref_index":5,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ","json":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ.json","graph_json":"https://pith.science/api/pith-number/42FNTVNGRGPDPGCLDZRLBGDJOZ/graph.json","events_json":"https://pith.science/api/pith-number/42FNTVNGRGPDPGCLDZRLBGDJOZ/events.json","paper":"https://pith.science/paper/42FNTVNG"},"agent_actions":{"view_html":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ","download_json":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ.json","view_paper":"https://pith.science/paper/42FNTVNG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2207.14255&json=true","fetch_graph":"https://pith.science/api/pith-number/42FNTVNGRGPDPGCLDZRLBGDJOZ/graph.json","fetch_events":"https://pith.science/api/pith-number/42FNTVNGRGPDPGCLDZRLBGDJOZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ/action/storage_attestation","attest_author":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ/action/author_attestation","sign_citation":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ/action/citation_signature","submit_replication":"https://pith.science/pith/42FNTVNGRGPDPGCLDZRLBGDJOZ/action/replication_record"}},"created_at":"2026-05-18T00:35:37.704085+00:00","updated_at":"2026-05-18T00:35:37.704085+00:00"}