{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:C5ZU7LGCFGBDSAEUMWA3A4TJLM","short_pith_number":"pith:C5ZU7LGC","schema_version":"1.0","canonical_sha256":"17734facc229823900946581b072695b15a68e361e32d1c6ac9f34d26b7f29ac","source":{"kind":"arxiv","id":"2405.14394","version":2},"attestation_state":"computed","paper":{"title":"Instruction Tuning With Loss Over Instructions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Adam X. Yang, Aldo Lipani, Bin Wu, Emine Yilmaz, Laurence Aitchison, Zhengyan Shi","submitted_at":"2024-05-23T10:12:03Z","abstract_excerpt":"Instruction tuning plays a crucial role in shaping the outputs of language models (LMs) to desired styles. In this work, we propose a simple yet effective method, Instruction Modelling (IM), which trains LMs by applying a loss function to the instruction and prompt part rather than solely to the output part. Through experiments across 21 diverse benchmarks, we show that, in many scenarios, IM can effectively improve the LM performance on both NLP tasks (e.g., MMLU, TruthfulQA, and HumanEval) and open-ended generation benchmarks (e.g., MT-Bench and AlpacaEval). Remarkably, in the most advantage"},"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":"2405.14394","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-05-23T10:12:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cf33f174ff25ba43a0b3f38141ce06583ec3eb38bf22a003de2ccf9300cf4070","abstract_canon_sha256":"216a62eafb20f0f315b8e3456ffef49782f63f2c3aaa5f80d5a28828e4af46b1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:14:54.338527Z","signature_b64":"MpqubaV4TP3zy75XGJ7+8fax1wTvhJagDSHqNfnu1y8Ovz1Cv3duEWepJzgLrhXHbV0XoebVQfiHuy8nOXfoAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"17734facc229823900946581b072695b15a68e361e32d1c6ac9f34d26b7f29ac","last_reissued_at":"2026-07-05T09:14:54.338012Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:14:54.338012Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Instruction Tuning With Loss Over Instructions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Adam X. Yang, Aldo Lipani, Bin Wu, Emine Yilmaz, Laurence Aitchison, Zhengyan Shi","submitted_at":"2024-05-23T10:12:03Z","abstract_excerpt":"Instruction tuning plays a crucial role in shaping the outputs of language models (LMs) to desired styles. In this work, we propose a simple yet effective method, Instruction Modelling (IM), which trains LMs by applying a loss function to the instruction and prompt part rather than solely to the output part. Through experiments across 21 diverse benchmarks, we show that, in many scenarios, IM can effectively improve the LM performance on both NLP tasks (e.g., MMLU, TruthfulQA, and HumanEval) and open-ended generation benchmarks (e.g., MT-Bench and AlpacaEval). Remarkably, in the most advantage"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.14394","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/2405.14394/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":"2405.14394","created_at":"2026-07-05T09:14:54.338075+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.14394v2","created_at":"2026-07-05T09:14:54.338075+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.14394","created_at":"2026-07-05T09:14:54.338075+00:00"},{"alias_kind":"pith_short_12","alias_value":"C5ZU7LGCFGBD","created_at":"2026-07-05T09:14:54.338075+00:00"},{"alias_kind":"pith_short_16","alias_value":"C5ZU7LGCFGBDSAEU","created_at":"2026-07-05T09:14:54.338075+00:00"},{"alias_kind":"pith_short_8","alias_value":"C5ZU7LGC","created_at":"2026-07-05T09:14:54.338075+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05304","citing_title":"What Should Agents Say? Action-state Communication for Efficient Multi-Agent Systems","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM","json":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM.json","graph_json":"https://pith.science/api/pith-number/C5ZU7LGCFGBDSAEUMWA3A4TJLM/graph.json","events_json":"https://pith.science/api/pith-number/C5ZU7LGCFGBDSAEUMWA3A4TJLM/events.json","paper":"https://pith.science/paper/C5ZU7LGC"},"agent_actions":{"view_html":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM","download_json":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM.json","view_paper":"https://pith.science/paper/C5ZU7LGC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.14394&json=true","fetch_graph":"https://pith.science/api/pith-number/C5ZU7LGCFGBDSAEUMWA3A4TJLM/graph.json","fetch_events":"https://pith.science/api/pith-number/C5ZU7LGCFGBDSAEUMWA3A4TJLM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM/action/storage_attestation","attest_author":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM/action/author_attestation","sign_citation":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM/action/citation_signature","submit_replication":"https://pith.science/pith/C5ZU7LGCFGBDSAEUMWA3A4TJLM/action/replication_record"}},"created_at":"2026-07-05T09:14:54.338075+00:00","updated_at":"2026-07-05T09:14:54.338075+00:00"}