{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:2CRB4BTXCM5KVUCZ6L6D3MCQUD","short_pith_number":"pith:2CRB4BTX","schema_version":"1.0","canonical_sha256":"d0a21e0677133aaad059f2fc3db050a0c26093565f25fc0bb2f1bd1061007af1","source":{"kind":"arxiv","id":"2407.13833","version":2},"attestation_state":"computed","paper":{"title":"Phi-3 Safety Post-Training: Aligning Language Models with a \"Break-Fix\" Cycle","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Alex Kessler, Amanda Minnich, Amit Garg, Andrew Berkley, Atabak Ashfaq, Blake Bullwinkel, Bolor-Erdene Jagdagdorj, Christian Seifert, Daniel Perez-Becker, David Majercak, Dongwoo Kim, Emman Haider, Gary Lopez, Hiteshi Sharma, Jianwen Zhang, Maggie Engler, Martin Pouliot, Nina Chikanov, Pete Bryan, Piyush Madan, Raja Sekhar Rao Dheekonda, Ram Shankar Siva Kumar, Richard Lundeen, Roman Lutz, Shahed Warreth, Shiven Chawla, Solianna Herrera, Thomas Portet, Tori Westerhoff, Wen Wen, Ziyi Yang","submitted_at":"2024-07-18T18:06:59Z","abstract_excerpt":"Recent innovations in language model training have demonstrated that it is possible to create highly performant models that are small enough to run on a smartphone. As these models are deployed in an increasing number of domains, it is critical to ensure that they are aligned with human preferences and safety considerations. In this report, we present our methodology for safety aligning the Phi-3 series of language models. We utilized a \"break-fix\" cycle, performing multiple rounds of dataset curation, safety post-training, benchmarking, red teaming, and vulnerability identification to cover a"},"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":"2407.13833","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-18T18:06:59Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"879c8fe5e01fd522bcaba5b2388353cd345ae437f959901f9bcfd6ac9f436fc2","abstract_canon_sha256":"2ab9cc7c6839894acf1a80f28c44be1af2743a6d1fe5cfab220229ee026049c5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:58:25.448146Z","signature_b64":"bOkq6TtIdpN7P5+RA9SuC8HZXM+YyKJ4sCxxWZkAvpttLdKtpmJ/Kwm6cATFOXD7nJxOMEBK0y1sq3rJNAl0Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d0a21e0677133aaad059f2fc3db050a0c26093565f25fc0bb2f1bd1061007af1","last_reissued_at":"2026-07-05T08:58:25.447658Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:58:25.447658Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Phi-3 Safety Post-Training: Aligning Language Models with a \"Break-Fix\" Cycle","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Alex Kessler, Amanda Minnich, Amit Garg, Andrew Berkley, Atabak Ashfaq, Blake Bullwinkel, Bolor-Erdene Jagdagdorj, Christian Seifert, Daniel Perez-Becker, David Majercak, Dongwoo Kim, Emman Haider, Gary Lopez, Hiteshi Sharma, Jianwen Zhang, Maggie Engler, Martin Pouliot, Nina Chikanov, Pete Bryan, Piyush Madan, Raja Sekhar Rao Dheekonda, Ram Shankar Siva Kumar, Richard Lundeen, Roman Lutz, Shahed Warreth, Shiven Chawla, Solianna Herrera, Thomas Portet, Tori Westerhoff, Wen Wen, Ziyi Yang","submitted_at":"2024-07-18T18:06:59Z","abstract_excerpt":"Recent innovations in language model training have demonstrated that it is possible to create highly performant models that are small enough to run on a smartphone. As these models are deployed in an increasing number of domains, it is critical to ensure that they are aligned with human preferences and safety considerations. In this report, we present our methodology for safety aligning the Phi-3 series of language models. We utilized a \"break-fix\" cycle, performing multiple rounds of dataset curation, safety post-training, benchmarking, red teaming, and vulnerability identification to cover a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.13833","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/2407.13833/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":"2407.13833","created_at":"2026-07-05T08:58:25.447715+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.13833v2","created_at":"2026-07-05T08:58:25.447715+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.13833","created_at":"2026-07-05T08:58:25.447715+00:00"},{"alias_kind":"pith_short_12","alias_value":"2CRB4BTXCM5K","created_at":"2026-07-05T08:58:25.447715+00:00"},{"alias_kind":"pith_short_16","alias_value":"2CRB4BTXCM5KVUCZ","created_at":"2026-07-05T08:58:25.447715+00:00"},{"alias_kind":"pith_short_8","alias_value":"2CRB4BTX","created_at":"2026-07-05T08:58:25.447715+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07535","citing_title":"Multilingual Refusal Alignment for Safer Large Language Models","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08565","citing_title":"EinSort: Sorting is All We Need for Tensorizing LLM","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29498","citing_title":"Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2509.26238","citing_title":"Beyond Linear Probes: Dynamic Safety Monitoring for Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04019","citing_title":"Redefining AI Red Teaming in the Agentic Era: From Weeks to Hours","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2503.01743","citing_title":"Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02860","citing_title":"Standing on the Shoulders of Giants: Stabilized Knowledge Distillation for Cross--Language Code Clone Detection","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD","json":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD.json","graph_json":"https://pith.science/api/pith-number/2CRB4BTXCM5KVUCZ6L6D3MCQUD/graph.json","events_json":"https://pith.science/api/pith-number/2CRB4BTXCM5KVUCZ6L6D3MCQUD/events.json","paper":"https://pith.science/paper/2CRB4BTX"},"agent_actions":{"view_html":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD","download_json":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD.json","view_paper":"https://pith.science/paper/2CRB4BTX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.13833&json=true","fetch_graph":"https://pith.science/api/pith-number/2CRB4BTXCM5KVUCZ6L6D3MCQUD/graph.json","fetch_events":"https://pith.science/api/pith-number/2CRB4BTXCM5KVUCZ6L6D3MCQUD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD/action/storage_attestation","attest_author":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD/action/author_attestation","sign_citation":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD/action/citation_signature","submit_replication":"https://pith.science/pith/2CRB4BTXCM5KVUCZ6L6D3MCQUD/action/replication_record"}},"created_at":"2026-07-05T08:58:25.447715+00:00","updated_at":"2026-07-05T08:58:25.447715+00:00"}