{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GM6SEGDPLQ5IXOTCWKUXOODNBR","short_pith_number":"pith:GM6SEGDP","schema_version":"1.0","canonical_sha256":"333d22186f5c3a8bba62b2a977386d0c5b3d90f931cc8074d9ddbfe15c49b048","source":{"kind":"arxiv","id":"2404.14618","version":1},"attestation_state":"computed","paper":{"title":"Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ahmed Hassan Awadallah, Ankur Mallick, Chi Wang, Dujian Ding, Laks V.S. Lakshmanan, Robert Sim, Subhabrata Mukherjee, Victor Ruhle","submitted_at":"2024-04-22T23:06:42Z","abstract_excerpt":"Large language models (LLMs) excel in most NLP tasks but also require expensive cloud servers for deployment due to their size, while smaller models that can be deployed on lower cost (e.g., edge) devices, tend to lag behind in terms of response quality. Therefore in this work we propose a hybrid inference approach which combines their respective strengths to save cost and maintain quality. Our approach uses a router that assigns queries to the small or large model based on the predicted query difficulty and the desired quality level. The desired quality level can be tuned dynamically at test "},"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":"2404.14618","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-22T23:06:42Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"c9b8872cbb2de58b8d7557181fec4a8c6ae29631ae5451135c9e703703c49d30","abstract_canon_sha256":"268802378f6fa4648fa52a15271df3c253ef69c436825e649e6a859eb97028e1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:11:07.745300Z","signature_b64":"Wn+W4XGAqcjVzlq2Y9WLHjt+3NY80E6VNnN5bp/miwfwhx8xfUUBnOlEaxilZqOgE4cJdlBCg/aUT5mPe2ldCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"333d22186f5c3a8bba62b2a977386d0c5b3d90f931cc8074d9ddbfe15c49b048","last_reissued_at":"2026-07-05T08:11:07.744778Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:11:07.744778Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ahmed Hassan Awadallah, Ankur Mallick, Chi Wang, Dujian Ding, Laks V.S. Lakshmanan, Robert Sim, Subhabrata Mukherjee, Victor Ruhle","submitted_at":"2024-04-22T23:06:42Z","abstract_excerpt":"Large language models (LLMs) excel in most NLP tasks but also require expensive cloud servers for deployment due to their size, while smaller models that can be deployed on lower cost (e.g., edge) devices, tend to lag behind in terms of response quality. Therefore in this work we propose a hybrid inference approach which combines their respective strengths to save cost and maintain quality. Our approach uses a router that assigns queries to the small or large model based on the predicted query difficulty and the desired quality level. The desired quality level can be tuned dynamically at test "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.14618","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/2404.14618/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":"2404.14618","created_at":"2026-07-05T08:11:07.744854+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.14618v1","created_at":"2026-07-05T08:11:07.744854+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.14618","created_at":"2026-07-05T08:11:07.744854+00:00"},{"alias_kind":"pith_short_12","alias_value":"GM6SEGDPLQ5I","created_at":"2026-07-05T08:11:07.744854+00:00"},{"alias_kind":"pith_short_16","alias_value":"GM6SEGDPLQ5IXOTC","created_at":"2026-07-05T08:11:07.744854+00:00"},{"alias_kind":"pith_short_8","alias_value":"GM6SEGDP","created_at":"2026-07-05T08:11:07.744854+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":30,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22902","citing_title":"Agent-as-a-Router: Agentic Model Routing for Coding Tasks","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20295","citing_title":"Token-Operations-Oriented Inference Optimization Techniques for Large Models","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31163","citing_title":"ComplianceGate: Classifier-Gated Multi-Tier LLM Routing for Inference in Regulated Industries","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00053","citing_title":"SWE-Router: Routing in Multi-turn Agentic Software Engineering Tasks","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04378","citing_title":"DLLG: Dynamic Logit-Level Gating of LLM Experts","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03800","citing_title":"Trading Human Curation for Synthetic Augmentation in RLVR","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31163","citing_title":"ComplianceGate: Classifier-Gated Multi-Tier LLM Routing for Inference in Regulated Industries","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22902","citing_title":"Agent-as-a-Router: Agentic Model Routing for Coding Tasks","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07587","citing_title":"The Routing Plateau: Understanding and Breaking the Accuracy Limits of LLM Routers","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30693","citing_title":"Triaging Threats to Specialized Guardrails","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2410.15761","citing_title":"Optimal Query Allocation in Extractive QA with LLMs: A Learning-to-Defer Framework with Theoretical Guarantees","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2505.12601","citing_title":"Rethinking Predictive Modeling for LLM Routing: When Simple kNN Beats Complex Learned Routers","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18796","citing_title":"UCCI: Calibrated Uncertainty for Cost-Optimal LLM Cascade Routing","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16604","citing_title":"R2V Agent: Teaching SLMs When to Ask for Help","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17106","citing_title":"HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19099","citing_title":"DecisionBench: A Benchmark for Emergent Delegation in Long-Horizon Agentic Workflows","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2509.24814","citing_title":"A Greedy PDE Router for Blending Neural Operators and Classical Methods","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2601.05110","citing_title":"GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2603.21354","citing_title":"The Workload-Router-Pool Architecture for LLM Inference Optimization: A Vision Paper from the vLLM Semantic Router Project","ref_index":109,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11301","citing_title":"LatentRouter: Can We Choose the Right Multimodal Model Before Seeing Its Answer?","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11317","citing_title":"SOMA: Efficient Multi-turn LLM Serving via Small Language Model","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08686","citing_title":"Iterative Critique-and-Routing Controller for Multi-Agent Systems with Heterogeneous LLMs","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06339","citing_title":"A Regime Theory of Controller Class Selection for LLM Action Decisions","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01111","citing_title":"When Less is Enough: Efficient Inference via Collaborative Reasoning","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00334","citing_title":"AgentFloor: How Far Up the tool use Ladder Can Small Open-Weight Models Go?","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR","json":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR.json","graph_json":"https://pith.science/api/pith-number/GM6SEGDPLQ5IXOTCWKUXOODNBR/graph.json","events_json":"https://pith.science/api/pith-number/GM6SEGDPLQ5IXOTCWKUXOODNBR/events.json","paper":"https://pith.science/paper/GM6SEGDP"},"agent_actions":{"view_html":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR","download_json":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR.json","view_paper":"https://pith.science/paper/GM6SEGDP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.14618&json=true","fetch_graph":"https://pith.science/api/pith-number/GM6SEGDPLQ5IXOTCWKUXOODNBR/graph.json","fetch_events":"https://pith.science/api/pith-number/GM6SEGDPLQ5IXOTCWKUXOODNBR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR/action/storage_attestation","attest_author":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR/action/author_attestation","sign_citation":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR/action/citation_signature","submit_replication":"https://pith.science/pith/GM6SEGDPLQ5IXOTCWKUXOODNBR/action/replication_record"}},"created_at":"2026-07-05T08:11:07.744854+00:00","updated_at":"2026-07-05T08:11:07.744854+00:00"}