{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EZPL3V7TKF64ALKS5XFZLDPIOE","short_pith_number":"pith:EZPL3V7T","schema_version":"1.0","canonical_sha256":"265ebdd7f3517dc02d52edcb958de8710d82d354ad3d56c4f72f36887ac02d33","source":{"kind":"arxiv","id":"2506.19140","version":1},"attestation_state":"computed","paper":{"title":"Command-V: Pasting LLM Behaviors via Activation Profiles","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alexander Robey, Ali Payani, Avi Schwarzschild, Barry Wang, Charles Fleming, Daphne Ippolito, Mingjie Sun","submitted_at":"2025-06-23T21:21:49Z","abstract_excerpt":"Retrofitting large language models (LLMs) with new behaviors typically requires full finetuning or distillation-costly steps that must be repeated for every architecture. In this work, we introduce Command-V, a backpropagation-free behavior transfer method that copies an existing residual activation adapter from a donor model and pastes its effect into a recipient model. Command-V profiles layer activations on a small prompt set, derives linear converters between corresponding layers, and applies the donor intervention in the recipient's activation space. This process does not require access t"},"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":"2506.19140","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-23T21:21:49Z","cross_cats_sorted":[],"title_canon_sha256":"8114245aac631cf83e28a94038464cf3ba7aa0c1b8eb089b1afe52ef96cc2a95","abstract_canon_sha256":"8ed4298bd27fa77102917a4abaf1ea672bbe19e5a1cd3bbd6549f73103508c4a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:26:09.846891Z","signature_b64":"grITD+nGD2dm+NQ75sMz9aFNBbGI4iqNm9Sh1LNdfHFLYUzPs/gQZaq1+rVM7J25sJJa4f+ZrUuaCUjFZH3dCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"265ebdd7f3517dc02d52edcb958de8710d82d354ad3d56c4f72f36887ac02d33","last_reissued_at":"2026-07-05T11:26:09.846456Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:26:09.846456Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Command-V: Pasting LLM Behaviors via Activation Profiles","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alexander Robey, Ali Payani, Avi Schwarzschild, Barry Wang, Charles Fleming, Daphne Ippolito, Mingjie Sun","submitted_at":"2025-06-23T21:21:49Z","abstract_excerpt":"Retrofitting large language models (LLMs) with new behaviors typically requires full finetuning or distillation-costly steps that must be repeated for every architecture. In this work, we introduce Command-V, a backpropagation-free behavior transfer method that copies an existing residual activation adapter from a donor model and pastes its effect into a recipient model. Command-V profiles layer activations on a small prompt set, derives linear converters between corresponding layers, and applies the donor intervention in the recipient's activation space. This process does not require access t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.19140","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/2506.19140/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":"2506.19140","created_at":"2026-07-05T11:26:09.846516+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.19140v1","created_at":"2026-07-05T11:26:09.846516+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.19140","created_at":"2026-07-05T11:26:09.846516+00:00"},{"alias_kind":"pith_short_12","alias_value":"EZPL3V7TKF64","created_at":"2026-07-05T11:26:09.846516+00:00"},{"alias_kind":"pith_short_16","alias_value":"EZPL3V7TKF64ALKS","created_at":"2026-07-05T11:26:09.846516+00:00"},{"alias_kind":"pith_short_8","alias_value":"EZPL3V7T","created_at":"2026-07-05T11:26:09.846516+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.14217","citing_title":"PreFT: Prefill-only finetuning for efficient inference","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE","json":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE.json","graph_json":"https://pith.science/api/pith-number/EZPL3V7TKF64ALKS5XFZLDPIOE/graph.json","events_json":"https://pith.science/api/pith-number/EZPL3V7TKF64ALKS5XFZLDPIOE/events.json","paper":"https://pith.science/paper/EZPL3V7T"},"agent_actions":{"view_html":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE","download_json":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE.json","view_paper":"https://pith.science/paper/EZPL3V7T","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.19140&json=true","fetch_graph":"https://pith.science/api/pith-number/EZPL3V7TKF64ALKS5XFZLDPIOE/graph.json","fetch_events":"https://pith.science/api/pith-number/EZPL3V7TKF64ALKS5XFZLDPIOE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE/action/storage_attestation","attest_author":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE/action/author_attestation","sign_citation":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE/action/citation_signature","submit_replication":"https://pith.science/pith/EZPL3V7TKF64ALKS5XFZLDPIOE/action/replication_record"}},"created_at":"2026-07-05T11:26:09.846516+00:00","updated_at":"2026-07-05T11:26:09.846516+00:00"}