{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TPIR6TCPPO7H2O7DJJMM2RWJRP","short_pith_number":"pith:TPIR6TCP","schema_version":"1.0","canonical_sha256":"9bd11f4c4f7bbe7d3be34a58cd46c98bf3f81fc93d9ddd993c25229e3c8dea1a","source":{"kind":"arxiv","id":"2508.09192","version":1},"attestation_state":"computed","paper":{"title":"Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chenkai Xu, Hao Zhang, Jiachun Jin, Xu Wang, Yijie Jin, Zhijie Deng","submitted_at":"2025-08-08T04:51:37Z","abstract_excerpt":"Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to autoregressive (AR) LLMs for text generation, with the potential to decode multiple tokens in a single iteration. However, none of the existing open-source dLLMs have achieved superior inference speed over AR LLMs of similar size. This paper breaks this barrier based on a simple and effective strategy named discrete diffusion forcing (D2F). D2F equips dLLMs with two key capabilities: (1) block-wise autoregressive generation to enable KV cache utilization; (2) prediction of following tokens without requiring comp"},"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":"2508.09192","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-08T04:51:37Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2130e09f2ebf759c90081584a298d04e7cb542d4874a18c4e53b0d12986e27f3","abstract_canon_sha256":"12dda377352a27325601d8ac3a073ed00e466fa2cfac3fc0829925d85c191565"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:53:09.696510Z","signature_b64":"BjF7BWEpUn1u5kRJRzYAfRTQI9W+dU5Hx1nNX1gesHHGRMief4cyOTgmLoIKrVN7kCntYCSe9NkRCpRCu5CWCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9bd11f4c4f7bbe7d3be34a58cd46c98bf3f81fc93d9ddd993c25229e3c8dea1a","last_reissued_at":"2026-07-05T11:53:09.695985Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:53:09.695985Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chenkai Xu, Hao Zhang, Jiachun Jin, Xu Wang, Yijie Jin, Zhijie Deng","submitted_at":"2025-08-08T04:51:37Z","abstract_excerpt":"Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to autoregressive (AR) LLMs for text generation, with the potential to decode multiple tokens in a single iteration. However, none of the existing open-source dLLMs have achieved superior inference speed over AR LLMs of similar size. This paper breaks this barrier based on a simple and effective strategy named discrete diffusion forcing (D2F). D2F equips dLLMs with two key capabilities: (1) block-wise autoregressive generation to enable KV cache utilization; (2) prediction of following tokens without requiring comp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.09192","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/2508.09192/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":"2508.09192","created_at":"2026-07-05T11:53:09.696042+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.09192v1","created_at":"2026-07-05T11:53:09.696042+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.09192","created_at":"2026-07-05T11:53:09.696042+00:00"},{"alias_kind":"pith_short_12","alias_value":"TPIR6TCPPO7H","created_at":"2026-07-05T11:53:09.696042+00:00"},{"alias_kind":"pith_short_16","alias_value":"TPIR6TCPPO7H2O7D","created_at":"2026-07-05T11:53:09.696042+00:00"},{"alias_kind":"pith_short_8","alias_value":"TPIR6TCP","created_at":"2026-07-05T11:53:09.696042+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":19,"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":66,"is_internal_anchor":true},{"citing_arxiv_id":"2606.10820","citing_title":"K-Forcing: Joint Next-K-Token Decoding via Push-Forward Language Modeling","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04486","citing_title":"Global Sketch-Based Watermarking for Diffusion Language Models","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19349","citing_title":"Where to Place the Query? Unveiling and Mitigating Positional Bias in In-Context Learning for Diffusion LLMs via Decoding Dynamics","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29215","citing_title":"Multi-Block Diffusion Language Models","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29215","citing_title":"Multi-Block Diffusion Language Models","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2602.08404","citing_title":"TEAM: Temporal-Spatial Consistency Guided Expert Activation for MoE Diffusion Language Model Acceleration","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20813","citing_title":"PulseCol: Periodically Refreshed Column-Sparse Attention for Accelerating Diffusion Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14530","citing_title":"Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":78,"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":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14530","citing_title":"Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2512.15745","citing_title":"LLaDA2.0: Scaling Up Diffusion Language Models to 100B","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10980","citing_title":"LEAP: Unlocking dLLM Parallelism via Lookahead Early-Convergence Token Detection","ref_index":14,"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":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":78,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06330","citing_title":"STDec: Spatio-Temporal Stability Guided Decoding for dLLMs","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP","json":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP.json","graph_json":"https://pith.science/api/pith-number/TPIR6TCPPO7H2O7DJJMM2RWJRP/graph.json","events_json":"https://pith.science/api/pith-number/TPIR6TCPPO7H2O7DJJMM2RWJRP/events.json","paper":"https://pith.science/paper/TPIR6TCP"},"agent_actions":{"view_html":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP","download_json":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP.json","view_paper":"https://pith.science/paper/TPIR6TCP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.09192&json=true","fetch_graph":"https://pith.science/api/pith-number/TPIR6TCPPO7H2O7DJJMM2RWJRP/graph.json","fetch_events":"https://pith.science/api/pith-number/TPIR6TCPPO7H2O7DJJMM2RWJRP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP/action/storage_attestation","attest_author":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP/action/author_attestation","sign_citation":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP/action/citation_signature","submit_replication":"https://pith.science/pith/TPIR6TCPPO7H2O7DJJMM2RWJRP/action/replication_record"}},"created_at":"2026-07-05T11:53:09.696042+00:00","updated_at":"2026-07-05T11:53:09.696042+00:00"}