{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:XKNCSZMSDP6ZQZXTYOG7JJZQQU","short_pith_number":"pith:XKNCSZMS","schema_version":"1.0","canonical_sha256":"ba9a2965921bfd9866f3c38df4a7308517c929054503639651c62ccc85dc3f08","source":{"kind":"arxiv","id":"2405.17741","version":1},"attestation_state":"computed","paper":{"title":"LoRA-Switch: Boosting the Efficiency of Dynamic LLM Adapters via System-Algorithm Co-design","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Linghe Kong, Qingfeng He, Qingtian Feng, Qiyang Li, Rui Kong, Weijun Wang, Xinyu Fang, Yazhu Dong, Yuanchun Li, Yunxin Liu","submitted_at":"2024-05-28T01:53:26Z","abstract_excerpt":"Recent literature has found that an effective method to customize or further improve large language models (LLMs) is to add dynamic adapters, such as low-rank adapters (LoRA) with Mixture-of-Experts (MoE) structures. Though such dynamic adapters incur modest computational complexity, they surprisingly lead to huge inference latency overhead, slowing down the decoding speed by 2.5+ times. In this paper, we analyze the fine-grained costs of the dynamic adapters and find that the fragmented CUDA kernel calls are the root cause. Therefore, we propose LoRA-Switch, a system-algorithm co-designed arc"},"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":"2405.17741","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-05-28T01:53:26Z","cross_cats_sorted":[],"title_canon_sha256":"e1c2ac49ad97f314d10790d2db4a9f7a0c7a19c431c679a014d98f1335b7f0d3","abstract_canon_sha256":"faa9a0297bc7f252a301293c2fe069d6198c36ac360931d795254fb3745d77ac"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:24:10.546138Z","signature_b64":"nSqozg8fgw4kOA2QJg4QzHuxu3Ff/9ZphLTIY9uUmUV9PfUQJwpsQBeaIedPjV/JFZGGpdwn/gZRFLRcVaR3Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ba9a2965921bfd9866f3c38df4a7308517c929054503639651c62ccc85dc3f08","last_reissued_at":"2026-07-05T08:24:10.545694Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:24:10.545694Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LoRA-Switch: Boosting the Efficiency of Dynamic LLM Adapters via System-Algorithm Co-design","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Linghe Kong, Qingfeng He, Qingtian Feng, Qiyang Li, Rui Kong, Weijun Wang, Xinyu Fang, Yazhu Dong, Yuanchun Li, Yunxin Liu","submitted_at":"2024-05-28T01:53:26Z","abstract_excerpt":"Recent literature has found that an effective method to customize or further improve large language models (LLMs) is to add dynamic adapters, such as low-rank adapters (LoRA) with Mixture-of-Experts (MoE) structures. Though such dynamic adapters incur modest computational complexity, they surprisingly lead to huge inference latency overhead, slowing down the decoding speed by 2.5+ times. In this paper, we analyze the fine-grained costs of the dynamic adapters and find that the fragmented CUDA kernel calls are the root cause. Therefore, we propose LoRA-Switch, a system-algorithm co-designed arc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.17741","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/2405.17741/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":"2405.17741","created_at":"2026-07-05T08:24:10.545756+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.17741v1","created_at":"2026-07-05T08:24:10.545756+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.17741","created_at":"2026-07-05T08:24:10.545756+00:00"},{"alias_kind":"pith_short_12","alias_value":"XKNCSZMSDP6Z","created_at":"2026-07-05T08:24:10.545756+00:00"},{"alias_kind":"pith_short_16","alias_value":"XKNCSZMSDP6ZQZXT","created_at":"2026-07-05T08:24:10.545756+00:00"},{"alias_kind":"pith_short_8","alias_value":"XKNCSZMS","created_at":"2026-07-05T08:24:10.545756+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05698","citing_title":"Rethinking LoRA Memory Through the Lens of KV Cache Compression","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10023","citing_title":"FREE-Switch: Frequency-based Dynamic LoRA Switch for Style Transfer","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07173","citing_title":"InfiniLoRA: Disaggregated Multi-LoRA Serving for Large Language Models","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU","json":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU.json","graph_json":"https://pith.science/api/pith-number/XKNCSZMSDP6ZQZXTYOG7JJZQQU/graph.json","events_json":"https://pith.science/api/pith-number/XKNCSZMSDP6ZQZXTYOG7JJZQQU/events.json","paper":"https://pith.science/paper/XKNCSZMS"},"agent_actions":{"view_html":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU","download_json":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU.json","view_paper":"https://pith.science/paper/XKNCSZMS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.17741&json=true","fetch_graph":"https://pith.science/api/pith-number/XKNCSZMSDP6ZQZXTYOG7JJZQQU/graph.json","fetch_events":"https://pith.science/api/pith-number/XKNCSZMSDP6ZQZXTYOG7JJZQQU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU/action/storage_attestation","attest_author":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU/action/author_attestation","sign_citation":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU/action/citation_signature","submit_replication":"https://pith.science/pith/XKNCSZMSDP6ZQZXTYOG7JJZQQU/action/replication_record"}},"created_at":"2026-07-05T08:24:10.545756+00:00","updated_at":"2026-07-05T08:24:10.545756+00:00"}