{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:A35AM6EABHLQKT3ZTYANKQJPQN","short_pith_number":"pith:A35AM6EA","schema_version":"1.0","canonical_sha256":"06fa06788009d7054f799e00d5412f836ca0927544fa44c74863baf5d51d68d3","source":{"kind":"arxiv","id":"2602.06036","version":2},"attestation_state":"computed","paper":{"title":"DFlash: Block Diffusion for Flash Speculative Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jian Chen, Yesheng Liang, Zhijian Liu","submitted_at":"2026-02-05T18:59:30Z","abstract_excerpt":"Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency and poor GPU utilization. Speculative decoding mitigates this bottleneck by using a fast draft model whose outputs are verified in parallel by the target LLM; however, existing methods still rely on autoregressive drafting, which remains sequential and limits practical speedups. Diffusion LLMs offer a promising alternative by enabling parallel generation, but current diffusion models typically underperform compared with autoregressive models. In t"},"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":"2602.06036","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-05T18:59:30Z","cross_cats_sorted":[],"title_canon_sha256":"0e8334084a7f240bb4426e8ac325e61b82231790084ea4bd100ac3de18d7e6c4","abstract_canon_sha256":"09bd5f9fb406c5da53efae0dfe7353b7eb95a304a727de7ed6e4703fa5705e71"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:05:03.617081Z","signature_b64":"Vc3VCBaicolz7sOcnq1b8tGBk4MdWABIHB/zpiX/xu7hq4FT5pO5jtK6RLO/r2NEdGxkoPWyx/dsvn8pDVT0DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06fa06788009d7054f799e00d5412f836ca0927544fa44c74863baf5d51d68d3","last_reissued_at":"2026-05-29T01:05:03.616090Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:05:03.616090Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DFlash: Block Diffusion for Flash Speculative Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jian Chen, Yesheng Liang, Zhijian Liu","submitted_at":"2026-02-05T18:59:30Z","abstract_excerpt":"Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency and poor GPU utilization. Speculative decoding mitigates this bottleneck by using a fast draft model whose outputs are verified in parallel by the target LLM; however, existing methods still rely on autoregressive drafting, which remains sequential and limits practical speedups. Diffusion LLMs offer a promising alternative by enabling parallel generation, but current diffusion models typically underperform compared with autoregressive models. In t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.06036","kind":"arxiv","version":2},"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/2602.06036/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":"2602.06036","created_at":"2026-05-29T01:05:03.616216+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.06036v2","created_at":"2026-05-29T01:05:03.616216+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.06036","created_at":"2026-05-29T01:05:03.616216+00:00"},{"alias_kind":"pith_short_12","alias_value":"A35AM6EABHLQ","created_at":"2026-05-29T01:05:03.616216+00:00"},{"alias_kind":"pith_short_16","alias_value":"A35AM6EABHLQKT3Z","created_at":"2026-05-29T01:05:03.616216+00:00"},{"alias_kind":"pith_short_8","alias_value":"A35AM6EA","created_at":"2026-05-29T01:05:03.616216+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":15,"sample":[{"citing_arxiv_id":"2605.09329","citing_title":"Test-Time Speculation","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18810","citing_title":"D-PACE: Dynamic Position-Aware Cross-Entropy for Parallel Speculative Drafting","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12825","citing_title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20104","citing_title":"Draft Less, Retrieve More: Hybrid Tree Construction for Speculative Decoding","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20022","citing_title":"FlexDraft: Flexible Speculative Decoding via Attention Tuning and Bonus-Guided Calibration","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12825","citing_title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2604.26412","citing_title":"When Hidden States Drift: Can KV Caches Rescue Long-Range Speculative Decoding?","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2604.26412","citing_title":"When Hidden States Drift: Can KV Caches Rescue Long-Range Speculative Decoding?","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08632","citing_title":"PARD-2: Target-Aligned Parallel Draft Model for Dual-Mode Speculative Decoding","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09329","citing_title":"Test-Time Speculation","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09992","citing_title":"Attention Drift: What Autoregressive Speculative Decoding Models Learn","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2604.12989","citing_title":"Accelerating Speculative Decoding with Block Diffusion Draft Trees","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09731","citing_title":"SMART: When is it Actually Worth Expanding a Speculative Tree?","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN","json":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN.json","graph_json":"https://pith.science/api/pith-number/A35AM6EABHLQKT3ZTYANKQJPQN/graph.json","events_json":"https://pith.science/api/pith-number/A35AM6EABHLQKT3ZTYANKQJPQN/events.json","paper":"https://pith.science/paper/A35AM6EA"},"agent_actions":{"view_html":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN","download_json":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN.json","view_paper":"https://pith.science/paper/A35AM6EA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.06036&json=true","fetch_graph":"https://pith.science/api/pith-number/A35AM6EABHLQKT3ZTYANKQJPQN/graph.json","fetch_events":"https://pith.science/api/pith-number/A35AM6EABHLQKT3ZTYANKQJPQN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN/action/storage_attestation","attest_author":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN/action/author_attestation","sign_citation":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN/action/citation_signature","submit_replication":"https://pith.science/pith/A35AM6EABHLQKT3ZTYANKQJPQN/action/replication_record"}},"created_at":"2026-05-29T01:05:03.616216+00:00","updated_at":"2026-05-29T01:05:03.616216+00:00"}