{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HPCYHLSY4N5IZ3T74VYH44OOCW","short_pith_number":"pith:HPCYHLSY","schema_version":"1.0","canonical_sha256":"3bc583ae58e37a8cee7fe5707e71ce1582f12dc5226a37219d4a6141e397c157","source":{"kind":"arxiv","id":"2504.02263","version":4},"attestation_state":"computed","paper":{"title":"MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Cesar A. Stuardo, Chao Jin, Dongyang Wang, Haibin Lin, Haoran Wei, Huaping Zhou, Jianxi Ye, Jianzhe Xiao, Lingjun Liu, Li-wen Chang, Peng Wu, Ruidong Zhu, Xiao Yu, Xin Jin, Xinlei Zhang, Xin Liu, Xinyi Zhang, Xuanzhe Liu, Yang Cheng, Ziheng Jiang","submitted_at":"2025-04-03T04:20:44Z","abstract_excerpt":"Mixture-of-Experts (MoE) showcases tremendous potential to scale large language models (LLMs) with enhanced performance and reduced computational complexity. However, its sparsely activated architecture shifts feed-forward networks (FFNs) from being compute-intensive to memory-intensive during inference, leading to substantially lower GPU utilization and increased operational costs.\n  We present MegaScale-Infer, an efficient and cost-effective system for serving large-scale MoE models. MegaScale-Infer disaggregates attention and FFN modules within each model layer, enabling independent scaling"},"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":"2504.02263","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2025-04-03T04:20:44Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"746867b3dbc00cedbffc7c44947609f22b964b6d9803d065f909b49fc21cefea","abstract_canon_sha256":"e6f0d28e9e3092841b19d9efdd7b6c40676f22ac3540bebd3e29342676059218"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:31.471918Z","signature_b64":"7xT+wC/U0Q+eR8bI9/iJpUfO9hnJETZatRDjbtwIqAcyfA1wTyT1FLNNHh4T/QWrbhWV8/XD5dsRkp1h96aNAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3bc583ae58e37a8cee7fe5707e71ce1582f12dc5226a37219d4a6141e397c157","last_reissued_at":"2026-07-05T11:43:31.471388Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:31.471388Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Cesar A. Stuardo, Chao Jin, Dongyang Wang, Haibin Lin, Haoran Wei, Huaping Zhou, Jianxi Ye, Jianzhe Xiao, Lingjun Liu, Li-wen Chang, Peng Wu, Ruidong Zhu, Xiao Yu, Xin Jin, Xinlei Zhang, Xin Liu, Xinyi Zhang, Xuanzhe Liu, Yang Cheng, Ziheng Jiang","submitted_at":"2025-04-03T04:20:44Z","abstract_excerpt":"Mixture-of-Experts (MoE) showcases tremendous potential to scale large language models (LLMs) with enhanced performance and reduced computational complexity. However, its sparsely activated architecture shifts feed-forward networks (FFNs) from being compute-intensive to memory-intensive during inference, leading to substantially lower GPU utilization and increased operational costs.\n  We present MegaScale-Infer, an efficient and cost-effective system for serving large-scale MoE models. MegaScale-Infer disaggregates attention and FFN modules within each model layer, enabling independent scaling"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.02263","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/2504.02263/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":"2504.02263","created_at":"2026-07-05T11:43:31.471457+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.02263v4","created_at":"2026-07-05T11:43:31.471457+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.02263","created_at":"2026-07-05T11:43:31.471457+00:00"},{"alias_kind":"pith_short_12","alias_value":"HPCYHLSY4N5I","created_at":"2026-07-05T11:43:31.471457+00:00"},{"alias_kind":"pith_short_16","alias_value":"HPCYHLSY4N5IZ3T7","created_at":"2026-07-05T11:43:31.471457+00:00"},{"alias_kind":"pith_short_8","alias_value":"HPCYHLSY","created_at":"2026-07-05T11:43:31.471457+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":19,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2607.05876","citing_title":"Think Before You Grid-Search: Floor-First Triage for LLM Serving","ref_index":49,"is_internal_anchor":true},{"citing_arxiv_id":"2607.06202","citing_title":"UBEP: Re-architecting Expert Parallelism Communication Library for Production Superpods","ref_index":58,"is_internal_anchor":true},{"citing_arxiv_id":"2607.05876","citing_title":"Think Before You Grid-Search: Floor-First Triage for LLM Serving","ref_index":49,"is_internal_anchor":true},{"citing_arxiv_id":"2607.06202","citing_title":"UBEP: Re-architecting Expert Parallelism Communication Library for Production Superpods","ref_index":58,"is_internal_anchor":true},{"citing_arxiv_id":"2607.00151","citing_title":"SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering","ref_index":88,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00735","citing_title":"ViBE: Co-Optimizing Workload Skew and Hardware Variability for MoE Serving","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29207","citing_title":"KernelFlume: Elastic Core-Attention Scaling for Agentic Long-Context Decoding","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28302","citing_title":"How Far Can Disaggregation Go? A Design-Space Exploration of Attention-FFN Disaggregation for Efficient MoE LLM Serving","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2511.09557","citing_title":"Understanding and Improving Communication Performance in Multi-node LLM Inference","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2512.19179","citing_title":"CascadeInfer: Length-Aware Scheduling of LLM Serving with Low Latency and Load Balancing","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21312","citing_title":"Frontier: Towards Comprehensive and Accurate LLM Inference Simulation","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2510.05497","citing_title":"Patterns behind Chaos: Forecasting Data Movement for Efficient Large-Scale MoE LLM Inference","ref_index":82,"is_internal_anchor":false},{"citing_arxiv_id":"2601.21351","citing_title":"Analytical Provisioning for Attention-FFN Disaggregated LLM Serving under Stochastic Workloads","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11005","citing_title":"DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe Parallelism","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11277","citing_title":"Sieve: Dynamic Expert-Aware PIM Acceleration for Evolving Mixture-of-Experts Models","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10670","citing_title":"Surviving Partial Rank Failures in Wide Expert-Parallel MoE Inference","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26103","citing_title":"AMMA: A Multi-Chiplet Memory-Centric Architecture for Low-Latency 1M Context Attention Serving","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01280","citing_title":"Position: LLM Serving Needs Mathematical Optimization and Algorithmic Foundations, Not Just Heuristics","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19241","citing_title":"UniEP: Unified Expert-Parallel MoE MegaKernel for LLM Training","ref_index":53,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW","json":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW.json","graph_json":"https://pith.science/api/pith-number/HPCYHLSY4N5IZ3T74VYH44OOCW/graph.json","events_json":"https://pith.science/api/pith-number/HPCYHLSY4N5IZ3T74VYH44OOCW/events.json","paper":"https://pith.science/paper/HPCYHLSY"},"agent_actions":{"view_html":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW","download_json":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW.json","view_paper":"https://pith.science/paper/HPCYHLSY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.02263&json=true","fetch_graph":"https://pith.science/api/pith-number/HPCYHLSY4N5IZ3T74VYH44OOCW/graph.json","fetch_events":"https://pith.science/api/pith-number/HPCYHLSY4N5IZ3T74VYH44OOCW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW/action/storage_attestation","attest_author":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW/action/author_attestation","sign_citation":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW/action/citation_signature","submit_replication":"https://pith.science/pith/HPCYHLSY4N5IZ3T74VYH44OOCW/action/replication_record"}},"created_at":"2026-07-05T11:43:31.471457+00:00","updated_at":"2026-07-05T11:43:31.471457+00:00"}