{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:FCZGIG6YM6MT5QELDJIKM7ZAA6","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":"c0d3d84724b127fce2caceeb33a5adc2fc84b887648672cb76874fab169bb4a0","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-03-16T19:37:08Z","title_canon_sha256":"e435c8655d84f9c3aa0afc24a8b8288f6ba2cdf4d13ca277893c96bab91d73a8"},"schema_version":"1.0","source":{"id":"2603.15854","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.15854","created_at":"2026-05-18T03:10:03Z"},{"alias_kind":"arxiv_version","alias_value":"2603.15854v2","created_at":"2026-05-18T03:10:03Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.15854","created_at":"2026-05-18T03:10:03Z"},{"alias_kind":"pith_short_12","alias_value":"FCZGIG6YM6MT","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"FCZGIG6YM6MT5QEL","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"FCZGIG6Y","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:1514ad7e5ee95c79190f6f07bafda60000515b61065a8dd14887b9253fdbc683","target":"graph","created_at":"2026-05-18T03:10:03Z","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":"FlashSampling is an exact sampling primitive that fuses sampling into the LM-head matmul and never materializes the logits tensor in HBM; in tensor-parallel decoding it replaces all-gather with streaming peer-to-peer writes, achieving kernel-level speedups and up to 10% reduction in time per output token."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The claim that argmax decomposes exactly over vocabulary partitions (and that the hierarchical factorization for grouped variants preserves exact categorical sampling) holds without numerical error or edge-case failure when tiles are processed independently on-chip."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"FlashSampling performs exact Gumbel-max sampling inside the LM-head matmul via on-chip tiling and hierarchical argmax reduction, delivering up to 10% faster token generation in vLLM on datacenter GPUs without approximation."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"FlashSampling fuses exact categorical sampling into the LM-head matrix multiply so the full logits tensor is never written to HBM."}],"snapshot_sha256":"ce3cba47136aac04afdbdcb5d0d61f83fc2176d9935ac901de94d3750e91ac7e"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Sampling from a categorical distribution is mathematically simple, but in large-vocabulary decoding, it often triggers extra memory traffic and extra kernels after the LM head. We present FlashSampling, an exact sampling primitive that fuses sampling into the LM-head matmul and never materializes the logits tensor in HBM. The method is simple: compute logits tile-by-tile on chip, add Gumbel noise, keep only one maximizer per row and per vocabulary tile, and finish with a small reduction over tiles. In tensor-parallel decoding, FlashSampling replaces the all-gather of logits with streaming peer","authors_text":"Mengdi Wang, Tomas Ruiz, Xuyang Shen, Yifan Zhang, Yiran Zhong, Zhen Qin","cross_cats":["cs.AI","cs.CL"],"headline":"FlashSampling fuses exact categorical sampling into the LM-head matrix multiply so the full logits tensor is never written to HBM.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-03-16T19:37:08Z","title":"FlashSampling: Fast and Memory-Efficient Exact Sampling"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.15854","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-15T09:57:45.197851Z","id":"3f38b0a9-aca2-46ac-a984-5ab58db9a9e7","model_set":{"reader":"grok-4.3"},"one_line_summary":"FlashSampling performs exact Gumbel-max sampling inside the LM-head matmul via on-chip tiling and hierarchical argmax reduction, delivering up to 10% faster token generation in vLLM on datacenter GPUs without approximation.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"FlashSampling fuses exact categorical sampling into the LM-head matrix multiply so the full logits tensor is never written to HBM.","strongest_claim":"FlashSampling is an exact sampling primitive that fuses sampling into the LM-head matmul and never materializes the logits tensor in HBM; in tensor-parallel decoding it replaces all-gather with streaming peer-to-peer writes, achieving kernel-level speedups and up to 10% reduction in time per output token.","weakest_assumption":"The claim that argmax decomposes exactly over vocabulary partitions (and that the hierarchical factorization for grouped variants preserves exact categorical sampling) holds without numerical error or edge-case failure when tiles are processed independently on-chip."}},"verdict_id":"3f38b0a9-aca2-46ac-a984-5ab58db9a9e7"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:4803b03f919fb99ca9ef47adbe7508f72c80e370dbc76d1ee7d030c90c432b4c","target":"record","created_at":"2026-05-18T03:10:03Z","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":"c0d3d84724b127fce2caceeb33a5adc2fc84b887648672cb76874fab169bb4a0","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-03-16T19:37:08Z","title_canon_sha256":"e435c8655d84f9c3aa0afc24a8b8288f6ba2cdf4d13ca277893c96bab91d73a8"},"schema_version":"1.0","source":{"id":"2603.15854","kind":"arxiv","version":2}},"canonical_sha256":"28b2641bd867993ec08b1a50a67f20079be979139708f2d2798dec5064f538c1","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"28b2641bd867993ec08b1a50a67f20079be979139708f2d2798dec5064f538c1","first_computed_at":"2026-05-18T03:10:03.435468Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:10:03.435468Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"jpjY0Qsapxu/7gKmGKs5H4E5u2lcprvJBNdL+CcL6noCRZPTDrK+bd7R/IathgPywCGXEGNAYCvp1rRPHHFVAw==","signature_status":"signed_v1","signed_at":"2026-05-18T03:10:03.436072Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.15854","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:4803b03f919fb99ca9ef47adbe7508f72c80e370dbc76d1ee7d030c90c432b4c","sha256:1514ad7e5ee95c79190f6f07bafda60000515b61065a8dd14887b9253fdbc683"],"state_sha256":"ba4458025d9639ae89d1df3cc3b5267231b2699cf3752b30c06b379012e4b896"}