{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BO4RFU3NZAGKJDLDVBQCLRYD7B","short_pith_number":"pith:BO4RFU3N","schema_version":"1.0","canonical_sha256":"0bb912d36dc80ca48d63a86025c703f847a46feccf3e11647a99eca0711eb298","source":{"kind":"arxiv","id":"2410.13846","version":3},"attestation_state":"computed","paper":{"title":"LightTransfer: Your Long-Context LLM is Secretly a Hybrid Model with Effortless Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chao Du, Cunxiao Du, Fengzhuo Zhang, Min Lin, Tianyu Pang, Wei Gao, Xuan Zhang","submitted_at":"2024-10-17T17:58:14Z","abstract_excerpt":"Scaling language models to handle longer contexts introduces substantial memory challenges due to the growing cost of key-value (KV) caches. Motivated by the efficiency gains of hybrid models and the broad availability of pretrained large transformer backbones, we explore transitioning transformer models into hybrid architectures for a more efficient generation. In this work, we propose LightTransfer, a lightweight method that transforms models such as LLaMA into hybrid variants. Our approach identifies lazy layers -- those focusing on recent or initial tokens -- and replaces their full attent"},"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":"2410.13846","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-10-17T17:58:14Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"4e58ef715442b43375a9f07caaed1127dc1e3f567fb47a4d2a673a220b996f34","abstract_canon_sha256":"338b97a12a62f477514bf65c46f3d996c1ec5ad2e9cf0c881ac17ea0e3f7a06a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:05.473542Z","signature_b64":"ysVRknZ997705/uT+XH1qxWWPrrUEM/FhcTbeLxRe4hp1BJsnXco6L5eaIuGLkkBpZl1Jyd5AR8/COqDGlysBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0bb912d36dc80ca48d63a86025c703f847a46feccf3e11647a99eca0711eb298","last_reissued_at":"2026-05-20T00:04:05.472698Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:05.472698Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LightTransfer: Your Long-Context LLM is Secretly a Hybrid Model with Effortless Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chao Du, Cunxiao Du, Fengzhuo Zhang, Min Lin, Tianyu Pang, Wei Gao, Xuan Zhang","submitted_at":"2024-10-17T17:58:14Z","abstract_excerpt":"Scaling language models to handle longer contexts introduces substantial memory challenges due to the growing cost of key-value (KV) caches. Motivated by the efficiency gains of hybrid models and the broad availability of pretrained large transformer backbones, we explore transitioning transformer models into hybrid architectures for a more efficient generation. In this work, we propose LightTransfer, a lightweight method that transforms models such as LLaMA into hybrid variants. Our approach identifies lazy layers -- those focusing on recent or initial tokens -- and replaces their full attent"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.13846","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/2410.13846/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":"2410.13846","created_at":"2026-05-20T00:04:05.472835+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.13846v3","created_at":"2026-05-20T00:04:05.472835+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.13846","created_at":"2026-05-20T00:04:05.472835+00:00"},{"alias_kind":"pith_short_12","alias_value":"BO4RFU3NZAGK","created_at":"2026-05-20T00:04:05.472835+00:00"},{"alias_kind":"pith_short_16","alias_value":"BO4RFU3NZAGKJDLD","created_at":"2026-05-20T00:04:05.472835+00:00"},{"alias_kind":"pith_short_8","alias_value":"BO4RFU3N","created_at":"2026-05-20T00:04:05.472835+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2510.00231","citing_title":"The Pitfalls of KV Cache Compression","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2502.13189","citing_title":"MoBA: Mixture of Block Attention for Long-Context LLMs","ref_index":48,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B","json":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B.json","graph_json":"https://pith.science/api/pith-number/BO4RFU3NZAGKJDLDVBQCLRYD7B/graph.json","events_json":"https://pith.science/api/pith-number/BO4RFU3NZAGKJDLDVBQCLRYD7B/events.json","paper":"https://pith.science/paper/BO4RFU3N"},"agent_actions":{"view_html":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B","download_json":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B.json","view_paper":"https://pith.science/paper/BO4RFU3N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.13846&json=true","fetch_graph":"https://pith.science/api/pith-number/BO4RFU3NZAGKJDLDVBQCLRYD7B/graph.json","fetch_events":"https://pith.science/api/pith-number/BO4RFU3NZAGKJDLDVBQCLRYD7B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B/action/storage_attestation","attest_author":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B/action/author_attestation","sign_citation":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B/action/citation_signature","submit_replication":"https://pith.science/pith/BO4RFU3NZAGKJDLDVBQCLRYD7B/action/replication_record"}},"created_at":"2026-05-20T00:04:05.472835+00:00","updated_at":"2026-05-20T00:04:05.472835+00:00"}