{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GTP3RAT2XIBXGXMA3THUE7IO23","short_pith_number":"pith:GTP3RAT2","schema_version":"1.0","canonical_sha256":"34dfb8827aba03735d80dccf427d0ed6d200f37a370322e153918b8fc8729cee","source":{"kind":"arxiv","id":"2401.14361","version":3},"attestation_state":"computed","paper":{"title":"MoE-Infinity: Efficient MoE Inference on Personal Machines with Sparsity-Aware Expert Cache","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.PF"],"primary_cat":"cs.LG","authors_text":"Leyang Xue, Luo Mai, Mahesh Marina, Yao Fu, Zhan Lu","submitted_at":"2024-01-25T18:07:50Z","abstract_excerpt":"This paper presents MoE-Infinity, an efficient MoE inference system designed for personal machines with limited GPU memory capacity. The key idea for MoE-Infinity is that on personal machines, which are often single-user environments, MoE-based LLMs typically operate with a batch size of one. In this setting, MoE models exhibit a high degree of activation sparsity, meaning a small number of experts are frequently reused in generating tokens during the decode phase. Leveraging this idea, we design a sparsity-aware expert cache, which can trace the sparse activation of experts during inference a"},"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":"2401.14361","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-01-25T18:07:50Z","cross_cats_sorted":["cs.PF"],"title_canon_sha256":"d6e5f0c52d751d32acd500f3de11e0978d767c13f6460f7bc95107d83df9fb1a","abstract_canon_sha256":"ee86c1298bc4029c886d846053e078160b531ee6b29728429348026c1cde3983"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:30:00.453363Z","signature_b64":"eCRNCQSC0CjP29PUrvUDIXHSzwuWuuN0FL7QUipICrU2ySmgNlV7ecVPyaedIt/+/BltmKwIBB9rEZgGFvGxBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"34dfb8827aba03735d80dccf427d0ed6d200f37a370322e153918b8fc8729cee","last_reissued_at":"2026-07-05T10:30:00.452818Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:30:00.452818Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MoE-Infinity: Efficient MoE Inference on Personal Machines with Sparsity-Aware Expert Cache","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.PF"],"primary_cat":"cs.LG","authors_text":"Leyang Xue, Luo Mai, Mahesh Marina, Yao Fu, Zhan Lu","submitted_at":"2024-01-25T18:07:50Z","abstract_excerpt":"This paper presents MoE-Infinity, an efficient MoE inference system designed for personal machines with limited GPU memory capacity. The key idea for MoE-Infinity is that on personal machines, which are often single-user environments, MoE-based LLMs typically operate with a batch size of one. In this setting, MoE models exhibit a high degree of activation sparsity, meaning a small number of experts are frequently reused in generating tokens during the decode phase. Leveraging this idea, we design a sparsity-aware expert cache, which can trace the sparse activation of experts during inference a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.14361","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/2401.14361/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":"2401.14361","created_at":"2026-07-05T10:30:00.452893+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.14361v3","created_at":"2026-07-05T10:30:00.452893+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.14361","created_at":"2026-07-05T10:30:00.452893+00:00"},{"alias_kind":"pith_short_12","alias_value":"GTP3RAT2XIBX","created_at":"2026-07-05T10:30:00.452893+00:00"},{"alias_kind":"pith_short_16","alias_value":"GTP3RAT2XIBXGXMA","created_at":"2026-07-05T10:30:00.452893+00:00"},{"alias_kind":"pith_short_8","alias_value":"GTP3RAT2","created_at":"2026-07-05T10:30:00.452893+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21428","citing_title":"Does Mixture-of-Experts Actually Help Inference on Consumer and Edge Hardware? An Empirical Study","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21868","citing_title":"WiSP: A Working-Set View of Mixture-of-Experts Serving on Extremely Low-Resource Hardware","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21712","citing_title":"BatchGen: An Architecture for Scalable and Efficient Batch Inference","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2601.21198","citing_title":"ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17889","citing_title":"CoX-MoE: Coalesced Expert Execution for High-Throughput MoE Inference with AMX-Enabled CPU-GPU Co-Execution","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20179","citing_title":"TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offload","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19481","citing_title":"C2CServe: Leveraging NVLink-C2C for Elastic Serverless LLM Serving on MIG","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2508.12851","citing_title":"Accelerating Edge Inference for Distributed MoE Models with Latency-Optimized Expert Placement","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2509.07379","citing_title":"DuoServe-MoE: Dual-Phase Expert Prefetch and Caching for LLM Inference QoS Assurance","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23638","citing_title":"LayerScope: Predictive Cross-Layer Scheduling for Efficient Multi-Batch MoE Inference on Legacy Servers","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02715","citing_title":"FluxMoE: Decoupling Expert Residency for High-Performance MoE Serving","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05899","citing_title":"VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14626","citing_title":"ELMoE-3D: Leveraging Intrinsic Elasticity of MoE for Hybrid-Bonding-Enabled Self-Speculative Decoding in On-Premises Serving","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17182","citing_title":"Layer-wise MoE Routing Locality under Shared-Prefix Code Generation: Token-Identity Decomposition and Compile-Equivalent Fork Redundancy","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18788","citing_title":"Efficient Mixture-of-Experts LLM Inference with Apple Silicon NPUs","ref_index":48,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23","json":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23.json","graph_json":"https://pith.science/api/pith-number/GTP3RAT2XIBXGXMA3THUE7IO23/graph.json","events_json":"https://pith.science/api/pith-number/GTP3RAT2XIBXGXMA3THUE7IO23/events.json","paper":"https://pith.science/paper/GTP3RAT2"},"agent_actions":{"view_html":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23","download_json":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23.json","view_paper":"https://pith.science/paper/GTP3RAT2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.14361&json=true","fetch_graph":"https://pith.science/api/pith-number/GTP3RAT2XIBXGXMA3THUE7IO23/graph.json","fetch_events":"https://pith.science/api/pith-number/GTP3RAT2XIBXGXMA3THUE7IO23/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23/action/storage_attestation","attest_author":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23/action/author_attestation","sign_citation":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23/action/citation_signature","submit_replication":"https://pith.science/pith/GTP3RAT2XIBXGXMA3THUE7IO23/action/replication_record"}},"created_at":"2026-07-05T10:30:00.452893+00:00","updated_at":"2026-07-05T10:30:00.452893+00:00"}