{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HIQIFA5URNEMIYIFV5YCG3EA3O","short_pith_number":"pith:HIQIFA5U","schema_version":"1.0","canonical_sha256":"3a208283b48b48c46105af70236c80db92c3814a33f852d725de204f8bb21513","source":{"kind":"arxiv","id":"2407.00066","version":4},"attestation_state":"computed","paper":{"title":"Compress then Serve: Serving Thousands of LoRA Adapters with Little Overhead","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.DC","authors_text":"Jiacheng Zhu, Justin Solomon, Kristjan Greenewald, Leshem Choshen, Mikhail Yurochkin, Onkar Bhardwaj, Rickard Br\\\"uel-Gabrielsson","submitted_at":"2024-06-17T15:21:35Z","abstract_excerpt":"Fine-tuning large language models (LLMs) with low-rank adaptations (LoRAs) has become common practice, often yielding numerous copies of the same LLM differing only in their LoRA updates. This paradigm presents challenges for systems that serve real-time responses to queries that each involve a different LoRA. Prior works optimize the design of such systems but still require continuous loading and offloading of LoRAs, as it is infeasible to store thousands of LoRAs in GPU memory. To mitigate this issue, we investigate the efficacy of compression when serving LoRAs. We propose a method for the "},"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":"2407.00066","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2024-06-17T15:21:35Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"324a325cbd21422fe1880be6e0b01c103ba28b0f443ec7b79696d35272d50c31","abstract_canon_sha256":"a4e7f7053b62d6409988cddb591e70854f9eacf941b219bf1f55b95887869a06"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:18.425342Z","signature_b64":"YrK0A0YPSrMFn1y2rkNprWWq7FzX920AEAgWKrWlzuns3SDM7ic00ErmcASkGjlrKzH9QCwZotfEoMEOOxyOBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a208283b48b48c46105af70236c80db92c3814a33f852d725de204f8bb21513","last_reissued_at":"2026-07-05T11:12:18.424774Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:18.424774Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Compress then Serve: Serving Thousands of LoRA Adapters with Little Overhead","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.DC","authors_text":"Jiacheng Zhu, Justin Solomon, Kristjan Greenewald, Leshem Choshen, Mikhail Yurochkin, Onkar Bhardwaj, Rickard Br\\\"uel-Gabrielsson","submitted_at":"2024-06-17T15:21:35Z","abstract_excerpt":"Fine-tuning large language models (LLMs) with low-rank adaptations (LoRAs) has become common practice, often yielding numerous copies of the same LLM differing only in their LoRA updates. This paradigm presents challenges for systems that serve real-time responses to queries that each involve a different LoRA. Prior works optimize the design of such systems but still require continuous loading and offloading of LoRAs, as it is infeasible to store thousands of LoRAs in GPU memory. To mitigate this issue, we investigate the efficacy of compression when serving LoRAs. We propose a method for the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.00066","kind":"arxiv","version":4},"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/2407.00066/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":"2407.00066","created_at":"2026-07-05T11:12:18.424853+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.00066v4","created_at":"2026-07-05T11:12:18.424853+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.00066","created_at":"2026-07-05T11:12:18.424853+00:00"},{"alias_kind":"pith_short_12","alias_value":"HIQIFA5URNEM","created_at":"2026-07-05T11:12:18.424853+00:00"},{"alias_kind":"pith_short_16","alias_value":"HIQIFA5URNEMIYIF","created_at":"2026-07-05T11:12:18.424853+00:00"},{"alias_kind":"pith_short_8","alias_value":"HIQIFA5U","created_at":"2026-07-05T11:12:18.424853+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05300","citing_title":"Learning Only What Valid Adapters Can Express: Subspace-Constrained Adaptation Against Fine-Tuning Poisoning","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01973","citing_title":"Learn-To-Learn on Arbitrary Textual Conditioning: A Hypernetwork-Driven Meta-Gated LLM","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13779","citing_title":"MinT: Managed Infrastructure for Training and Serving Millions of LLMs","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13779","citing_title":"MinT: Managed Infrastructure for Training and Serving Millions of LLMs","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01973","citing_title":"Learn-To-Learn on Arbitrary Textual Conditioning: A Hypernetwork-Driven Meta-Gated LLM","ref_index":117,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O","json":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O.json","graph_json":"https://pith.science/api/pith-number/HIQIFA5URNEMIYIFV5YCG3EA3O/graph.json","events_json":"https://pith.science/api/pith-number/HIQIFA5URNEMIYIFV5YCG3EA3O/events.json","paper":"https://pith.science/paper/HIQIFA5U"},"agent_actions":{"view_html":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O","download_json":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O.json","view_paper":"https://pith.science/paper/HIQIFA5U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.00066&json=true","fetch_graph":"https://pith.science/api/pith-number/HIQIFA5URNEMIYIFV5YCG3EA3O/graph.json","fetch_events":"https://pith.science/api/pith-number/HIQIFA5URNEMIYIFV5YCG3EA3O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O/action/storage_attestation","attest_author":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O/action/author_attestation","sign_citation":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O/action/citation_signature","submit_replication":"https://pith.science/pith/HIQIFA5URNEMIYIFV5YCG3EA3O/action/replication_record"}},"created_at":"2026-07-05T11:12:18.424853+00:00","updated_at":"2026-07-05T11:12:18.424853+00:00"}