{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LH5LN2VC5IF5MWQLNE4MJPBQDQ","short_pith_number":"pith:LH5LN2VC","schema_version":"1.0","canonical_sha256":"59fab6eaa2ea0bd65a0b6938c4bc301c1d5bf8d4e8185c4ae287f8187656f5c4","source":{"kind":"arxiv","id":"2506.00204","version":1},"attestation_state":"computed","paper":{"title":"Structure-Aware Fill-in-the-Middle Pretraining for Code","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.SE"],"primary_cat":"cs.CL","authors_text":"Alvin Cheung, Linyuan Gong, Mostafa Elhoushi, Sida Wang","submitted_at":"2025-05-30T20:19:39Z","abstract_excerpt":"Fill-in-the-Middle (FIM) is a common pretraining method for code LLMs, where models complete code segments given surrounding context. However, existing LLMs treat code as plain text and mask random character spans. We propose and evaluate AST-FIM, a pretraining strategy that leverages Abstract Syntax Trees (ASTs) to mask complete syntactic structures at scale, ensuring coherent training examples better aligned with universal code structures and common code editing patterns such as blocks, expressions, or functions. To evaluate real-world fill-in-the-middle (FIM) programming tasks, we introduce"},"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":"2506.00204","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-30T20:19:39Z","cross_cats_sorted":["cs.AI","cs.SE"],"title_canon_sha256":"2a710f9919b8e84b5b296d12b82352641815159bc9146762c928f38e8d882d29","abstract_canon_sha256":"196e0e395e074298e6a9bf727e8a86253636d5ddd556c6488a390d2e9b0fe391"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:30.732941Z","signature_b64":"ff6kFdibh5FGLkQXG4bMh+58GJJcW7dq11pz26WWlWKgillyCob7093A474NM0zqoa+HOq6/3kawOqieW1IdAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"59fab6eaa2ea0bd65a0b6938c4bc301c1d5bf8d4e8185c4ae287f8187656f5c4","last_reissued_at":"2026-07-05T11:13:30.732462Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:30.732462Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Structure-Aware Fill-in-the-Middle Pretraining for Code","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.SE"],"primary_cat":"cs.CL","authors_text":"Alvin Cheung, Linyuan Gong, Mostafa Elhoushi, Sida Wang","submitted_at":"2025-05-30T20:19:39Z","abstract_excerpt":"Fill-in-the-Middle (FIM) is a common pretraining method for code LLMs, where models complete code segments given surrounding context. However, existing LLMs treat code as plain text and mask random character spans. We propose and evaluate AST-FIM, a pretraining strategy that leverages Abstract Syntax Trees (ASTs) to mask complete syntactic structures at scale, ensuring coherent training examples better aligned with universal code structures and common code editing patterns such as blocks, expressions, or functions. To evaluate real-world fill-in-the-middle (FIM) programming tasks, we introduce"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.00204","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/2506.00204/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":"2506.00204","created_at":"2026-07-05T11:13:30.732522+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.00204v1","created_at":"2026-07-05T11:13:30.732522+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.00204","created_at":"2026-07-05T11:13:30.732522+00:00"},{"alias_kind":"pith_short_12","alias_value":"LH5LN2VC5IF5","created_at":"2026-07-05T11:13:30.732522+00:00"},{"alias_kind":"pith_short_16","alias_value":"LH5LN2VC5IF5MWQL","created_at":"2026-07-05T11:13:30.732522+00:00"},{"alias_kind":"pith_short_8","alias_value":"LH5LN2VC","created_at":"2026-07-05T11:13:30.732522+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25198","citing_title":"Hide to Guide: Learning via Semantic Masking","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03130","citing_title":"Synthetic Hallucinations, Real Gains: Hard Negatives from Frontier Models for FIM Hallucination Mitigation","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ","json":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ.json","graph_json":"https://pith.science/api/pith-number/LH5LN2VC5IF5MWQLNE4MJPBQDQ/graph.json","events_json":"https://pith.science/api/pith-number/LH5LN2VC5IF5MWQLNE4MJPBQDQ/events.json","paper":"https://pith.science/paper/LH5LN2VC"},"agent_actions":{"view_html":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ","download_json":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ.json","view_paper":"https://pith.science/paper/LH5LN2VC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.00204&json=true","fetch_graph":"https://pith.science/api/pith-number/LH5LN2VC5IF5MWQLNE4MJPBQDQ/graph.json","fetch_events":"https://pith.science/api/pith-number/LH5LN2VC5IF5MWQLNE4MJPBQDQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ/action/storage_attestation","attest_author":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ/action/author_attestation","sign_citation":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ/action/citation_signature","submit_replication":"https://pith.science/pith/LH5LN2VC5IF5MWQLNE4MJPBQDQ/action/replication_record"}},"created_at":"2026-07-05T11:13:30.732522+00:00","updated_at":"2026-07-05T11:13:30.732522+00:00"}