{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:SGRHL77BNAP25FGDUODECICWAT","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"bf1f9ca812600cc95e0be27a7a4414028f4b550c704c5c74d72538c538b50d81","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-03T03:10:24Z","title_canon_sha256":"8f2d740a74c55c27cdcf58160f8ba9005e393ea49b602d7e6f62e60564e6bf54"},"schema_version":"1.0","source":{"id":"2605.02960","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.02960","created_at":"2026-05-20T00:00:40Z"},{"alias_kind":"arxiv_version","alias_value":"2605.02960v2","created_at":"2026-05-20T00:00:40Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.02960","created_at":"2026-05-20T00:00:40Z"},{"alias_kind":"pith_short_12","alias_value":"SGRHL77BNAP2","created_at":"2026-05-20T00:00:40Z"},{"alias_kind":"pith_short_16","alias_value":"SGRHL77BNAP25FGD","created_at":"2026-05-20T00:00:40Z"},{"alias_kind":"pith_short_8","alias_value":"SGRHL77B","created_at":"2026-05-20T00:00:40Z"}],"graph_snapshots":[{"event_id":"sha256:7890143c0c1e21abd607977868a4b64c036defe4fe6571695f3b0bd2bbca06e3","target":"graph","created_at":"2026-05-20T00:00:40Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"On Qwen3-235B-A22B across four hardware/precision configurations, MoE-Prefill delivers 1.35-1.37x throughput over the strongest distributed baseline on real-world workloads and up to 1.59x on long-context synthetic workloads, sustaining 29.8-36.2% per-GPU model FLOPs utilization."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The long, compute-bound forward passes of large-batch prefill open a per-layer window wide enough to stream expert weights in the background, replacing per-layer activation AllToAll with asynchronous weight AllGather fully overlapped with computation without new bottlenecks or accuracy loss."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MoE-Prefill achieves 1.35-1.59x higher throughput for prefill-only MoE serving by using asynchronous expert parallelism to overlap weight AllGather with computation and prefix-aware routing with true-FLOPs tracking."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"MoE prefill serving eliminates redundant overheads by asynchronously gathering expert weights during compute-bound phases."}],"snapshot_sha256":"74b8371a6a6832046d254f75f767c3ebd917729fcbae67b252d77215f264f953"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T17:01:31.756960Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.02960/integrity.json","findings":[],"snapshot_sha256":"a1878eac2e9beaf64465e0bee886c01093de3f9ce883ca40f498e2da03b942f6","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Production LLM workloads increasingly serve discriminative tasks, such as classification, recommendation, and verification, whose answers are read from the logits of a single prefill pass with no autoregressive decoding. Serving these prefill-only workloads on mixture-of-experts (MoE) models is bottlenecked not by compute but by the distributed execution required to fit the model: existing parallel strategies (tensor, expert, and pipeline parallelism) trade memory pressure for redundant computation, communication, and synchronization, severely degrading MoE prefill serving efficiency. We obser","authors_text":"Aurick Qiao, Juncheng Yang, Karthik Ganesan, Olatunji Ruwase, Samyam Rajbhandari, Yue Cheng, Yuxiong He, Zhaoyuan Su","cross_cats":[],"headline":"MoE prefill serving eliminates redundant overheads by asynchronously gathering expert weights during compute-bound phases.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-03T03:10:24Z","title":"MoE-Prefill: Zero Redundancy Overheads in MoE Prefill Serving"},"references":{"count":75,"internal_anchors":16,"resolved_work":75,"sample":[{"cited_arxiv_id":"2508.10925","doi":"","is_internal_anchor":true,"ref_index":1,"title":"gpt-oss-120b & gpt-oss-20b Model Card","work_id":"178c1f7e-4f19-4392-a45d-45a6dfa88ead","year":2025},{"cited_arxiv_id":"2308.16369","doi":"","is_internal_anchor":true,"ref_index":2,"title":"SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills","work_id":"3dbdd757-ca01-436f-acfd-12ffcd6f64c6","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Deepspeed-inference: enabling efficient in- ference of transformer models at unprecedented scale","work_id":"461bf7aa-e1b9-4309-a540-e2905d823c31","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"A survey on mixture of experts in large language models.IEEE Transactions on Knowledge and Data Engineering, 2025","work_id":"c76fa220-c8af-40e4-a69f-ee8ff33e469b","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Moe-lightning: High-throughput moe inference on memory-constrained gpus","work_id":"c7c2193d-2846-4a86-8642-116ade28549b","year":2025}],"snapshot_sha256":"e667420ecd0ad681b43542be402886428e8f63d7873ba26aa2adae8ae08a04b0"},"source":{"id":"2605.02960","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-19T17:38:54.747167Z","id":"83754aca-5224-4f93-b30f-082eafe82638","model_set":{"reader":"grok-4.3"},"one_line_summary":"MoE-Prefill achieves 1.35-1.59x higher throughput for prefill-only MoE serving by using asynchronous expert parallelism to overlap weight AllGather with computation and prefix-aware routing with true-FLOPs tracking.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"MoE prefill serving eliminates redundant overheads by asynchronously gathering expert weights during compute-bound phases.","strongest_claim":"On Qwen3-235B-A22B across four hardware/precision configurations, MoE-Prefill delivers 1.35-1.37x throughput over the strongest distributed baseline on real-world workloads and up to 1.59x on long-context synthetic workloads, sustaining 29.8-36.2% per-GPU model FLOPs utilization.","weakest_assumption":"The long, compute-bound forward passes of large-batch prefill open a per-layer window wide enough to stream expert weights in the background, replacing per-layer activation AllToAll with asynchronous weight AllGather fully overlapped with computation without new bottlenecks or accuracy loss."}},"verdict_id":"83754aca-5224-4f93-b30f-082eafe82638"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:8303b25519228d465f58715311784d5187811e61231238d6a998379f41e67422","target":"record","created_at":"2026-05-20T00:00:40Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"bf1f9ca812600cc95e0be27a7a4414028f4b550c704c5c74d72538c538b50d81","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-03T03:10:24Z","title_canon_sha256":"8f2d740a74c55c27cdcf58160f8ba9005e393ea49b602d7e6f62e60564e6bf54"},"schema_version":"1.0","source":{"id":"2605.02960","kind":"arxiv","version":2}},"canonical_sha256":"91a275ffe1681fae94c3a38641205604f0c2f8207ac4823a618be80750689201","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"91a275ffe1681fae94c3a38641205604f0c2f8207ac4823a618be80750689201","first_computed_at":"2026-05-20T00:00:40.477498Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:00:40.477498Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"YQEMyM8sArO+sU2UNbVzYk5GDQaVllId1AZkhparJvI3X0J1LF9/7bicYaLQsx0KX0wWZtyu2wzWNwjElGBHAA==","signature_status":"signed_v1","signed_at":"2026-05-20T00:00:40.478275Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.02960","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8303b25519228d465f58715311784d5187811e61231238d6a998379f41e67422","sha256:7890143c0c1e21abd607977868a4b64c036defe4fe6571695f3b0bd2bbca06e3"],"state_sha256":"7bf03a5554840b4f7fd0ccfb4d99f03bc65426aa2332c5211a2e09964f6b83c0"}