{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:S2HT6JMPE6EJFHYIOC5MLY757P","short_pith_number":"pith:S2HT6JMP","schema_version":"1.0","canonical_sha256":"968f3f258f2788929f0870bac5e3fdfbe1cc74171d0c94ec105959d507e2acc7","source":{"kind":"arxiv","id":"2505.19893","version":1},"attestation_state":"computed","paper":{"title":"ESLM: Risk-Averse Selective Language Modeling for Efficient Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Melis Ilayda Bal, Michael Muehlebach, Volkan Cevher","submitted_at":"2025-05-26T12:23:26Z","abstract_excerpt":"Large language model pretraining is compute-intensive, yet many tokens contribute marginally to learning, resulting in inefficiency. We introduce Efficient Selective Language Modeling (ESLM), a risk-aware algorithm that improves training efficiency and distributional robustness by performing online token-level batch selection. ESLM leverages per-token statistics (e.g., entropy or loss) and applies value-at-risk thresholding to retain only the most informative tokens per batch. This data-centric mechanism reshapes the training loss, prioritizing high-risk tokens and eliminating redundant gradie"},"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":"2505.19893","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-26T12:23:26Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"283e93c2b269f37f0d34aeb04f1754f7f049de5899bc5675c43fa3f07a148691","abstract_canon_sha256":"84443c770689619c652869503566e6a8516d78ff83389a010597ac711381aab4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:49.569534Z","signature_b64":"YPBEGj9HVuwyLDcDJEAgR3ZUIq6WDIJ6IxhxbD7aifjQATlLWQkTY1P+BagZBpQkJF0GTT4oJ7PGxFs2Iru3Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"968f3f258f2788929f0870bac5e3fdfbe1cc74171d0c94ec105959d507e2acc7","last_reissued_at":"2026-07-05T11:09:49.569024Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:49.569024Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ESLM: Risk-Averse Selective Language Modeling for Efficient Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Melis Ilayda Bal, Michael Muehlebach, Volkan Cevher","submitted_at":"2025-05-26T12:23:26Z","abstract_excerpt":"Large language model pretraining is compute-intensive, yet many tokens contribute marginally to learning, resulting in inefficiency. We introduce Efficient Selective Language Modeling (ESLM), a risk-aware algorithm that improves training efficiency and distributional robustness by performing online token-level batch selection. ESLM leverages per-token statistics (e.g., entropy or loss) and applies value-at-risk thresholding to retain only the most informative tokens per batch. This data-centric mechanism reshapes the training loss, prioritizing high-risk tokens and eliminating redundant gradie"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.19893","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/2505.19893/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":"2505.19893","created_at":"2026-07-05T11:09:49.569077+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19893v1","created_at":"2026-07-05T11:09:49.569077+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19893","created_at":"2026-07-05T11:09:49.569077+00:00"},{"alias_kind":"pith_short_12","alias_value":"S2HT6JMPE6EJ","created_at":"2026-07-05T11:09:49.569077+00:00"},{"alias_kind":"pith_short_16","alias_value":"S2HT6JMPE6EJFHYI","created_at":"2026-07-05T11:09:49.569077+00:00"},{"alias_kind":"pith_short_8","alias_value":"S2HT6JMP","created_at":"2026-07-05T11:09:49.569077+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/S2HT6JMPE6EJFHYIOC5MLY757P","json":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P.json","graph_json":"https://pith.science/api/pith-number/S2HT6JMPE6EJFHYIOC5MLY757P/graph.json","events_json":"https://pith.science/api/pith-number/S2HT6JMPE6EJFHYIOC5MLY757P/events.json","paper":"https://pith.science/paper/S2HT6JMP"},"agent_actions":{"view_html":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P","download_json":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P.json","view_paper":"https://pith.science/paper/S2HT6JMP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19893&json=true","fetch_graph":"https://pith.science/api/pith-number/S2HT6JMPE6EJFHYIOC5MLY757P/graph.json","fetch_events":"https://pith.science/api/pith-number/S2HT6JMPE6EJFHYIOC5MLY757P/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P/action/timestamp_anchor","attest_storage":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P/action/storage_attestation","attest_author":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P/action/author_attestation","sign_citation":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P/action/citation_signature","submit_replication":"https://pith.science/pith/S2HT6JMPE6EJFHYIOC5MLY757P/action/replication_record"}},"created_at":"2026-07-05T11:09:49.569077+00:00","updated_at":"2026-07-05T11:09:49.569077+00:00"}