{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FLYCOVZC5XQDVUSWFKYZDGCP5H","short_pith_number":"pith:FLYCOVZC","schema_version":"1.0","canonical_sha256":"2af0275722ede03ad2562ab191984fe9e62d319bb56f0ea24c0ff1d392a3d360","source":{"kind":"arxiv","id":"2505.15781","version":1},"attestation_state":"computed","paper":{"title":"dKV-Cache: The Cache for Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Gongfan Fang, Runpeng Yu, Xinchao Wang, Xinyin Ma","submitted_at":"2025-05-21T17:32:10Z","abstract_excerpt":"Diffusion Language Models (DLMs) have been seen as a promising competitor for autoregressive language models. However, diffusion language models have long been constrained by slow inference. A core challenge is that their non-autoregressive architecture and bidirectional attention preclude the key-value cache that accelerates decoding. We address this bottleneck by proposing a KV-cache-like mechanism, delayed KV-Cache, for the denoising process of DLMs. Our approach is motivated by the observation that different tokens have distinct representation dynamics throughout the diffusion process. Acc"},"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":"2505.15781","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-21T17:32:10Z","cross_cats_sorted":[],"title_canon_sha256":"8ff88400e9e7ac20167c5be46b8a18bbcf30175b1d2f567686c8f8a992302572","abstract_canon_sha256":"8f5400f0512ae844be449c56d8ad4748d776e075b33711f52ec101c41963d264"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:06:57.955958Z","signature_b64":"EGXLK1xvGk3T4EtvoWAI611RmA3jT4RLozwfoy/tsoFjrZT/PiLfDCmbtbc5I9WOtP1jdjGVG0PjD8Iub7b4DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2af0275722ede03ad2562ab191984fe9e62d319bb56f0ea24c0ff1d392a3d360","last_reissued_at":"2026-07-05T11:06:57.955481Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:06:57.955481Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"dKV-Cache: The Cache for Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Gongfan Fang, Runpeng Yu, Xinchao Wang, Xinyin Ma","submitted_at":"2025-05-21T17:32:10Z","abstract_excerpt":"Diffusion Language Models (DLMs) have been seen as a promising competitor for autoregressive language models. However, diffusion language models have long been constrained by slow inference. A core challenge is that their non-autoregressive architecture and bidirectional attention preclude the key-value cache that accelerates decoding. We address this bottleneck by proposing a KV-cache-like mechanism, delayed KV-Cache, for the denoising process of DLMs. Our approach is motivated by the observation that different tokens have distinct representation dynamics throughout the diffusion process. Acc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15781","kind":"arxiv","version":1},"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/2505.15781/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":"2505.15781","created_at":"2026-07-05T11:06:57.955542+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15781v1","created_at":"2026-07-05T11:06:57.955542+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15781","created_at":"2026-07-05T11:06:57.955542+00:00"},{"alias_kind":"pith_short_12","alias_value":"FLYCOVZC5XQD","created_at":"2026-07-05T11:06:57.955542+00:00"},{"alias_kind":"pith_short_16","alias_value":"FLYCOVZC5XQDVUSW","created_at":"2026-07-05T11:06:57.955542+00:00"},{"alias_kind":"pith_short_8","alias_value":"FLYCOVZC","created_at":"2026-07-05T11:06:57.955542+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":28,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05722","citing_title":"Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding","ref_index":62,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25331","citing_title":"Improved Large Language Diffusion Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01775","citing_title":"Set Diffusion: Interpolating Token Orderings Between Autoregression and Diffusion for Fast and Flexible Decoding","ref_index":139,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08411","citing_title":"AsyncLane: Decoupling Refinement from Advancement in Diffusion Language Model Decoding","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02544","citing_title":"SimSD: Simple Speculative Decoding in Diffusion Language Models","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29814","citing_title":"Nemotron-Labs-Diffusion-Image: Advancing Masked Discrete Diffusion for High-Resolution Image Synthesis","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26120","citing_title":"Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00724","citing_title":"WaveFilter: Enhancing the Long-Context Capability of Diffusion LLMs via Wavelet-Guided KV Cache Filtering","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20813","citing_title":"PulseCol: Periodically Refreshed Column-Sparse Attention for Accelerating Diffusion Language Models","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12825","citing_title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18165","citing_title":"Elastic-dLLM: Position Preserving Context Compression and Augmentation of Diffusion LLMs","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19470","citing_title":"Drifting Objectives for Refining Discrete Diffusion Language Models","ref_index":29,"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":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2505.23606","citing_title":"Muddit: Liberating Generation Beyond Text-to-Image with a Unified Discrete Diffusion Model","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2508.19982","citing_title":"Diffusion Language Models Know the Answer Before Decoding","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2510.04525","citing_title":"Demystifying MaskGIT Sampler and Beyond: Adaptive Order Selection in Masked Diffusion","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2512.14067","citing_title":"Efficient-DLM: From Autoregressive to Diffusion Language Models, and Beyond in Speed","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2603.07475","citing_title":"A Comparative analysis of Layer-wise Representational Capacity in AR and Diffusion LLMs","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12825","citing_title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09999","citing_title":"Muninn: Your Trajectory Diffusion Model But Faster","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09536","citing_title":"TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08134","citing_title":"DARE: Diffusion Language Model Activation Reuse for Efficient Inference","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00161","citing_title":"Consistent Diffusion Language Models","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18995","citing_title":"$R^2$-dLLM: Accelerating Diffusion Large Language Models via Spatio-Temporal Redundancy Reduction","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H","json":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H.json","graph_json":"https://pith.science/api/pith-number/FLYCOVZC5XQDVUSWFKYZDGCP5H/graph.json","events_json":"https://pith.science/api/pith-number/FLYCOVZC5XQDVUSWFKYZDGCP5H/events.json","paper":"https://pith.science/paper/FLYCOVZC"},"agent_actions":{"view_html":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H","download_json":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H.json","view_paper":"https://pith.science/paper/FLYCOVZC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15781&json=true","fetch_graph":"https://pith.science/api/pith-number/FLYCOVZC5XQDVUSWFKYZDGCP5H/graph.json","fetch_events":"https://pith.science/api/pith-number/FLYCOVZC5XQDVUSWFKYZDGCP5H/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H/action/storage_attestation","attest_author":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H/action/author_attestation","sign_citation":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H/action/citation_signature","submit_replication":"https://pith.science/pith/FLYCOVZC5XQDVUSWFKYZDGCP5H/action/replication_record"}},"created_at":"2026-07-05T11:06:57.955542+00:00","updated_at":"2026-07-05T11:06:57.955542+00:00"}