{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:HMRV444CW7DNSSM7UWMDXRPECS","short_pith_number":"pith:HMRV444C","canonical_record":{"source":{"id":"2412.11041","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-15T03:58:38Z","cross_cats_sorted":[],"title_canon_sha256":"d0beede6dc3c4b5655fe630738e2e8459067671901da2b5dc84c9fdcf914c279","abstract_canon_sha256":"153e575e650661b3fab8ad6e2d2ab26948b1e00ce6491f7ef7d0c18fb799e6f3"},"schema_version":"1.0"},"canonical_sha256":"3b235e7382b7c6d9499fa5983bc5e41485c195892550658e9c962474d965c98a","source":{"kind":"arxiv","id":"2412.11041","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2412.11041","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"arxiv_version","alias_value":"2412.11041v3","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.11041","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"pith_short_12","alias_value":"HMRV444CW7DN","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"pith_short_16","alias_value":"HMRV444CW7DNSSM7","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"pith_short_8","alias_value":"HMRV444C","created_at":"2026-07-05T11:09:05Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:HMRV444CW7DNSSM7UWMDXRPECS","target":"record","payload":{"canonical_record":{"source":{"id":"2412.11041","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-15T03:58:38Z","cross_cats_sorted":[],"title_canon_sha256":"d0beede6dc3c4b5655fe630738e2e8459067671901da2b5dc84c9fdcf914c279","abstract_canon_sha256":"153e575e650661b3fab8ad6e2d2ab26948b1e00ce6491f7ef7d0c18fb799e6f3"},"schema_version":"1.0"},"canonical_sha256":"3b235e7382b7c6d9499fa5983bc5e41485c195892550658e9c962474d965c98a","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:05.377544Z","signature_b64":"WcmZuSzM/MQHhXePaxHDGzTydj+n/+aHFPsL/38xg+l7rsOJ7Ut0e6Xeq/S0Z9yyAO/EgPChqdkD7Usbuc+6AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3b235e7382b7c6d9499fa5983bc5e41485c195892550658e9c962474d965c98a","last_reissued_at":"2026-07-05T11:09:05.377094Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:05.377094Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2412.11041","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:09:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"9lpeoSGhYTwnd2IfUNeObRURMgB1oyDKw2sin34hMvdgdiBQc332PsX8hVXuFMCW4IpLnb+bwaoN+2Zf0OwwBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-16T23:41:14.028417Z"},"content_sha256":"fc40f0d277b628b85d46ad2c32b0dc4d03399dcf500ead9c18ab8fecdf0e98a8","schema_version":"1.0","event_id":"sha256:fc40f0d277b628b85d46ad2c32b0dc4d03399dcf500ead9c18ab8fecdf0e98a8"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:HMRV444CW7DNSSM7UWMDXRPECS","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Separate the Wheat from the Chaff: A Post-Hoc Approach to Safety Re-Alignment for Fine-Tuned Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bing Qin, Di Wu, Xin Lu, Yanyan Zhao","submitted_at":"2024-12-15T03:58:38Z","abstract_excerpt":"Although large language models (LLMs) achieve effective safety alignment at the time of release, they still face various safety challenges. A key issue is that fine-tuning often compromises the safety alignment of LLMs. To address this issue, we propose a method named IRR (Identify, Remove, and Recalibrate for Safety Realignment) that performs safety realignment for LLMs. The core of IRR is to identify and remove unsafe delta parameters from the fine-tuned models, while recalibrating the retained ones. We evaluate the effectiveness of IRR across various datasets, including both full fine-tunin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.11041","kind":"arxiv","version":3},"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/2412.11041/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:09:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Imc2PgQNOlOtf5E4jkNdvgqTZmO9TNs2L2cYDgoA8lUYbHvniedSgKsCl4NQZJPnJS1ibxJI3obrpZPvgXr0AQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-16T23:41:14.028794Z"},"content_sha256":"ae5f060b50a870c667cbdcf0bec6c40ddce9d5b39fd61c6d132f3fbd871a1275","schema_version":"1.0","event_id":"sha256:ae5f060b50a870c667cbdcf0bec6c40ddce9d5b39fd61c6d132f3fbd871a1275"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/HMRV444CW7DNSSM7UWMDXRPECS/bundle.json","state_url":"https://pith.science/pith/HMRV444CW7DNSSM7UWMDXRPECS/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/HMRV444CW7DNSSM7UWMDXRPECS/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-16T23:41:14Z","links":{"resolver":"https://pith.science/pith/HMRV444CW7DNSSM7UWMDXRPECS","bundle":"https://pith.science/pith/HMRV444CW7DNSSM7UWMDXRPECS/bundle.json","state":"https://pith.science/pith/HMRV444CW7DNSSM7UWMDXRPECS/state.json","well_known_bundle":"https://pith.science/.well-known/pith/HMRV444CW7DNSSM7UWMDXRPECS/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:HMRV444CW7DNSSM7UWMDXRPECS","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"153e575e650661b3fab8ad6e2d2ab26948b1e00ce6491f7ef7d0c18fb799e6f3","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-15T03:58:38Z","title_canon_sha256":"d0beede6dc3c4b5655fe630738e2e8459067671901da2b5dc84c9fdcf914c279"},"schema_version":"1.0","source":{"id":"2412.11041","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2412.11041","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"arxiv_version","alias_value":"2412.11041v3","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.11041","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"pith_short_12","alias_value":"HMRV444CW7DN","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"pith_short_16","alias_value":"HMRV444CW7DNSSM7","created_at":"2026-07-05T11:09:05Z"},{"alias_kind":"pith_short_8","alias_value":"HMRV444C","created_at":"2026-07-05T11:09:05Z"}],"graph_snapshots":[{"event_id":"sha256:ae5f060b50a870c667cbdcf0bec6c40ddce9d5b39fd61c6d132f3fbd871a1275","target":"graph","created_at":"2026-07-05T11:09:05Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2412.11041/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Although large language models (LLMs) achieve effective safety alignment at the time of release, they still face various safety challenges. A key issue is that fine-tuning often compromises the safety alignment of LLMs. To address this issue, we propose a method named IRR (Identify, Remove, and Recalibrate for Safety Realignment) that performs safety realignment for LLMs. The core of IRR is to identify and remove unsafe delta parameters from the fine-tuned models, while recalibrating the retained ones. We evaluate the effectiveness of IRR across various datasets, including both full fine-tunin","authors_text":"Bing Qin, Di Wu, Xin Lu, Yanyan Zhao","cross_cats":[],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-15T03:58:38Z","title":"Separate the Wheat from the Chaff: A Post-Hoc Approach to Safety Re-Alignment for Fine-Tuned Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.11041","kind":"arxiv","version":3},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:fc40f0d277b628b85d46ad2c32b0dc4d03399dcf500ead9c18ab8fecdf0e98a8","target":"record","created_at":"2026-07-05T11:09:05Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"153e575e650661b3fab8ad6e2d2ab26948b1e00ce6491f7ef7d0c18fb799e6f3","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-15T03:58:38Z","title_canon_sha256":"d0beede6dc3c4b5655fe630738e2e8459067671901da2b5dc84c9fdcf914c279"},"schema_version":"1.0","source":{"id":"2412.11041","kind":"arxiv","version":3}},"canonical_sha256":"3b235e7382b7c6d9499fa5983bc5e41485c195892550658e9c962474d965c98a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"3b235e7382b7c6d9499fa5983bc5e41485c195892550658e9c962474d965c98a","first_computed_at":"2026-07-05T11:09:05.377094Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:09:05.377094Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"WcmZuSzM/MQHhXePaxHDGzTydj+n/+aHFPsL/38xg+l7rsOJ7Ut0e6Xeq/S0Z9yyAO/EgPChqdkD7Usbuc+6AA==","signature_status":"signed_v1","signed_at":"2026-07-05T11:09:05.377544Z","signed_message":"canonical_sha256_bytes"},"source_id":"2412.11041","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:fc40f0d277b628b85d46ad2c32b0dc4d03399dcf500ead9c18ab8fecdf0e98a8","sha256:ae5f060b50a870c667cbdcf0bec6c40ddce9d5b39fd61c6d132f3fbd871a1275"],"state_sha256":"8c72010589b3f640611987658bbe7fb1f74a1b22c5c43c6340f3b7419625c61d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"zJdGItwSpNYmqZGfS3UwflqQrJHiqYJ5m+napBXkJ/E/KT6iKslQKnh32rmrOSMbg1sN/nhSnhkfE9vqMGmUBg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-16T23:41:14.031242Z","bundle_sha256":"6a1dd897e04bea85cf3f815160fbf36a3b2e3554d42a73ea9c4abcf324094c84"}}