{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ZW67IC5TVTEVM75APPZUYYMUZL","short_pith_number":"pith:ZW67IC5T","schema_version":"1.0","canonical_sha256":"cdbdf40bb3acc9567fa07bf34c6194cac65e7a7edf1940d6d865ac759c9a4607","source":{"kind":"arxiv","id":"2505.22618","version":3},"attestation_state":"computed","paper":{"title":"Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Diffusion LLMs can reach up to 27 times higher throughput by adding a reusable block-wise KV cache and decoding only high-confidence tokens in parallel.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chengyue Wu, Enze Xie, Hao Zhang, Ligeng Zhu, Ping Luo, Shizhe Diao, Shuchen Xue, Song Han, Zhijian Liu","submitted_at":"2025-05-28T17:39:15Z","abstract_excerpt":"Diffusion-based large language models (Diffusion LLMs) have shown promise for non-autoregressive text generation with parallel decoding capabilities. However, the practical inference speed of open-sourced Diffusion LLMs often lags behind autoregressive models due to the lack of Key-Value (KV) Cache and quality degradation when decoding multiple tokens simultaneously. To bridge this gap, we introduce a novel block-wise approximate KV Cache mechanism tailored for bidirectional diffusion models, enabling cache reuse with negligible performance drop. Additionally, we identify the root cause of gen"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2505.22618","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-28T17:39:15Z","cross_cats_sorted":[],"title_canon_sha256":"1d06872d1f805a26fa9fccdb0e1669fc666f60fd1dd3785d349b212af66fdf32","abstract_canon_sha256":"e44c8d1bf337f28a9eef6d07f8920fbb29ab4a8cc9ed4ba42ac5d50a4bbe688a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:49.114586Z","signature_b64":"HGJkjoPa92YnH+CMVS+NVDvdiOQRbsWOMyMujYN8D2pWzh5fQZ+tTRm/X7wc5y0JSB6cGvp5fphvBHQZ1tKRDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cdbdf40bb3acc9567fa07bf34c6194cac65e7a7edf1940d6d865ac759c9a4607","last_reissued_at":"2026-05-17T23:38:49.113794Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:49.113794Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Diffusion LLMs can reach up to 27 times higher throughput by adding a reusable block-wise KV cache and decoding only high-confidence tokens in parallel.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chengyue Wu, Enze Xie, Hao Zhang, Ligeng Zhu, Ping Luo, Shizhe Diao, Shuchen Xue, Song Han, Zhijian Liu","submitted_at":"2025-05-28T17:39:15Z","abstract_excerpt":"Diffusion-based large language models (Diffusion LLMs) have shown promise for non-autoregressive text generation with parallel decoding capabilities. However, the practical inference speed of open-sourced Diffusion LLMs often lags behind autoregressive models due to the lack of Key-Value (KV) Cache and quality degradation when decoding multiple tokens simultaneously. To bridge this gap, we introduce a novel block-wise approximate KV Cache mechanism tailored for bidirectional diffusion models, enabling cache reuse with negligible performance drop. Additionally, we identify the root cause of gen"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experimental results on LLaDA and Dream models across multiple LLM benchmarks demonstrate up to 27.6× throughput improvement with minimal accuracy loss, closing the performance gap with autoregressive models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the block-wise approximate KV cache introduces only negligible performance drop and that a single confidence threshold can be chosen to preserve generation quality across benchmarks without post-hoc per-task retuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Fast-dLLM adds reusable KV cache blocks and selective parallel decoding to diffusion LLMs, closing most of the speed gap with autoregressive models without retraining.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Diffusion LLMs can reach up to 27 times higher throughput by adding a reusable block-wise KV cache and decoding only high-confidence tokens in parallel.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f24406c2f002b1a64e35976b50236fbcc4633fac5ca074731963f35337f28e30"},"source":{"id":"2505.22618","kind":"arxiv","version":3},"verdict":{"id":"e45f5b97-466a-4ff5-bba7-06dba8cc791c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T04:22:39.565036Z","strongest_claim":"Experimental results on LLaDA and Dream models across multiple LLM benchmarks demonstrate up to 27.6× throughput improvement with minimal accuracy loss, closing the performance gap with autoregressive models.","one_line_summary":"Fast-dLLM adds reusable KV cache blocks and selective parallel decoding to diffusion LLMs, closing most of the speed gap with autoregressive models without retraining.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the block-wise approximate KV cache introduces only negligible performance drop and that a single confidence threshold can be chosen to preserve generation quality across benchmarks without post-hoc per-task retuning.","pith_extraction_headline":"Diffusion LLMs can reach up to 27 times higher throughput by adding a reusable block-wise KV cache and decoding only high-confidence tokens in parallel."},"references":{"count":44,"sample":[{"doi":"","year":2025,"title":"Chiu, Zhihan Yang, Zhixuan Qi, Jiaqi Han, Subham Sekhar Sahoo, and V olodymyr Kuleshov","work_id":"a7d122b4-332a-4b4e-b20f-261ba8497fa0","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Structured denoising diffusion models in discrete state-spaces","work_id":"73976884-b26f-4171-a159-b0f0a8b9ce9e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"A continuous time framework for discrete denoising models","work_id":"f67f008a-e2de-44f8-a5a0-6dd58dc2befd","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Fast sampling via de-randomization for discrete diffusion models","work_id":"19c1e430-1409-412d-b13e-f87d1e5714a7","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Discrete flow matching","work_id":"d3810acb-2472-492b-b3b2-2cefdf738e9c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":44,"snapshot_sha256":"01b65f53997ee7bc50ebae18ef6441d398d39a42222c7c90bd024da7edd75c84","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d77a8a9c1f6ad21895b1a43885db0c963b200bcfefb761bd05082ee3cdeef337"},"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":"2505.22618","created_at":"2026-05-17T23:38:49.113911+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.22618v3","created_at":"2026-05-17T23:38:49.113911+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.22618","created_at":"2026-05-17T23:38:49.113911+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZW67IC5TVTEV","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZW67IC5TVTEVM75A","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZW67IC5T","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":44,"internal_anchor_count":44,"sample":[{"citing_arxiv_id":"2605.22967","citing_title":"Learned Relay Representations for Forward-Thinking Discrete Diffusion Models","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.23163","citing_title":"Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2511.18801","citing_title":"PartDiffuser: Part-wise 3D Mesh Generation via Discrete Diffusion","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2601.21484","citing_title":"ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2602.16813","citing_title":"Flow Map Language Models: One-step Language Modeling via Continuous Denoising","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20199","citing_title":"FlowLM: Few-Step Language Modeling via Diffusion-to-Flow Adaptation","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":50,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20813","citing_title":"PulseCol: Periodically Refreshed Column-Sparse Attention for Accelerating Diffusion Language Models","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11854","citing_title":"Self-Distilled Trajectory-Aware Boltzmann Modeling: Bridging the Training-Inference Discrepancy in Diffusion Language Models","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15676","citing_title":"Dynamic Chunking for Diffusion Language Models","ref_index":43,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18165","citing_title":"Elastic-dLLM: Position Preserving Context Compression and Augmentation of Diffusion LLMs","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19660","citing_title":"OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyond","ref_index":67,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19470","citing_title":"Drifting Objectives for Refining Discrete Diffusion Language Models","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20022","citing_title":"FlexDraft: Flexible Speculative Decoding via Attention Tuning and Bonus-Guided Calibration","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16941","citing_title":"Roll Out and Roll Back: Diffusion LLMs are Their Own Efficiency Teachers","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":84,"is_internal_anchor":true},{"citing_arxiv_id":"2509.20624","citing_title":"FS-DFM: Fast and Accurate Long Text Generation with Few-Step Diffusion Language Models","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2510.06133","citing_title":"CreditDecoding: Accelerating Parallel Decoding in Diffusion Large Language Models with Trace Credit","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2510.09477","citing_title":"Efficient Autoregressive Inference for Transformer Probabilistic Models","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2510.13117","citing_title":"On the Reasoning Abilities of Masked Diffusion Language Models","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2510.18165","citing_title":"Saber: An Efficient Sampling with Adaptive Acceleration and Backtracking Enhanced Remasking for Diffusion Language Model","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2512.14067","citing_title":"Efficient-DLM: From Autoregressive to Diffusion Language Models, and Beyond in Speed","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2601.20706","citing_title":"NPU Design for Diffusion Language Model Inference","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2601.21484","citing_title":"ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2602.16813","citing_title":"Flow Map Language Models: One-step Language Modeling via Continuous Denoising","ref_index":10,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL","json":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL.json","graph_json":"https://pith.science/api/pith-number/ZW67IC5TVTEVM75APPZUYYMUZL/graph.json","events_json":"https://pith.science/api/pith-number/ZW67IC5TVTEVM75APPZUYYMUZL/events.json","paper":"https://pith.science/paper/ZW67IC5T"},"agent_actions":{"view_html":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL","download_json":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL.json","view_paper":"https://pith.science/paper/ZW67IC5T","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.22618&json=true","fetch_graph":"https://pith.science/api/pith-number/ZW67IC5TVTEVM75APPZUYYMUZL/graph.json","fetch_events":"https://pith.science/api/pith-number/ZW67IC5TVTEVM75APPZUYYMUZL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL/action/storage_attestation","attest_author":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL/action/author_attestation","sign_citation":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL/action/citation_signature","submit_replication":"https://pith.science/pith/ZW67IC5TVTEVM75APPZUYYMUZL/action/replication_record"}},"created_at":"2026-05-17T23:38:49.113911+00:00","updated_at":"2026-05-17T23:38:49.113911+00:00"}