{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BTT26H53S5CHOTCCLFFKY2PSXC","short_pith_number":"pith:BTT26H53","schema_version":"1.0","canonical_sha256":"0ce7af1fbb9744774c42594aac69f2b8aaceb5070c5b144a53b2c43179d12287","source":{"kind":"arxiv","id":"2503.09573","version":3},"attestation_state":"computed","paper":{"title":"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Block diffusion language models interpolate between autoregressive and diffusion approaches to support arbitrary-length generation.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aaron Gokaslan, Jiaqi Han, Justin T. Chiu, Marianne Arriola, Subham Sekhar Sahoo, Volodymyr Kuleshov, Zhihan Yang, Zhixuan Qi","submitted_at":"2025-03-12T17:43:40Z","abstract_excerpt":"Diffusion language models offer unique benefits over autoregressive models due to their potential for parallelized generation and controllability, yet they lag in likelihood modeling and are limited to fixed-length generation. In this work, we introduce a class of block diffusion language models that interpolate between discrete denoising diffusion and autoregressive models. Block diffusion overcomes key limitations of both approaches by supporting flexible-length generation and improving inference efficiency with KV caching and parallel token sampling. We propose a recipe for building effecti"},"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":true},"canonical_record":{"source":{"id":"2503.09573","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-12T17:43:40Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9b40e1aacce6ca5e6225bfb376e1eeb9416502edb782bc9350072db557806efa","abstract_canon_sha256":"ee4131df89e827e58acc2c2dd5262bbc9d93f6c405a3ea3d7dd13344728ed9ab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:52.723027Z","signature_b64":"uNgYRic0Ih/Nx0GXZM9xvo+5L8WsJ5X4IiSwxxhOVywFbaX5g/VTvUzR/Y919Zj/7mdAx3Q62qiwyABzEMUYAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0ce7af1fbb9744774c42594aac69f2b8aaceb5070c5b144a53b2c43179d12287","last_reissued_at":"2026-05-17T23:38:52.722376Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:52.722376Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Block diffusion language models interpolate between autoregressive and diffusion approaches to support arbitrary-length generation.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aaron Gokaslan, Jiaqi Han, Justin T. Chiu, Marianne Arriola, Subham Sekhar Sahoo, Volodymyr Kuleshov, Zhihan Yang, Zhixuan Qi","submitted_at":"2025-03-12T17:43:40Z","abstract_excerpt":"Diffusion language models offer unique benefits over autoregressive models due to their potential for parallelized generation and controllability, yet they lag in likelihood modeling and are limited to fixed-length generation. In this work, we introduce a class of block diffusion language models that interpolate between discrete denoising diffusion and autoregressive models. Block diffusion overcomes key limitations of both approaches by supporting flexible-length generation and improving inference efficiency with KV caching and parallel token sampling. We propose a recipe for building effecti"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Block diffusion sets a new state-of-the-art performance among diffusion models on language modeling benchmarks and enables generation of arbitrary-length sequences.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the proposed training algorithm, estimators of gradient variance, and data-driven noise schedules will produce stable and effective models that generalize beyond the training data without hidden instabilities or overfitting.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Block diffusion language models interpolate between autoregressive and diffusion models to support flexible-length generation and achieve state-of-the-art performance among diffusion models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Block diffusion language models interpolate between autoregressive and diffusion approaches to support arbitrary-length generation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d8d2c38f29f42a19534692b11e3fbc296219ddb7aa85975d74a2d57a4b7032c9"},"source":{"id":"2503.09573","kind":"arxiv","version":3},"verdict":{"id":"de29bb41-2ce2-426c-89e5-052e5d87aada","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T10:53:58.246957Z","strongest_claim":"Block diffusion sets a new state-of-the-art performance among diffusion models on language modeling benchmarks and enables generation of arbitrary-length sequences.","one_line_summary":"Block diffusion language models interpolate between autoregressive and diffusion models to support flexible-length generation and achieve state-of-the-art performance among diffusion models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the proposed training algorithm, estimators of gradient variance, and data-driven noise schedules will produce stable and effective models that generalize beyond the training data without hidden instabilities or overfitting.","pith_extraction_headline":"Block diffusion language models interpolate between autoregressive and diffusion approaches to support arbitrary-length generation."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"83b5fdca3750b24655294c819078d4126181615e16a6e578dcf11a9646ec8cb7"},"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":"2503.09573","created_at":"2026-05-17T23:38:52.722484+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.09573v3","created_at":"2026-05-17T23:38:52.722484+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.09573","created_at":"2026-05-17T23:38:52.722484+00:00"},{"alias_kind":"pith_short_12","alias_value":"BTT26H53S5CH","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"BTT26H53S5CHOTCC","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"BTT26H53","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":39,"internal_anchor_count":39,"sample":[{"citing_arxiv_id":"2605.22967","citing_title":"Learned Relay Representations for Forward-Thinking Discrete Diffusion Models","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2605.23163","citing_title":"Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2601.10348","citing_title":"Training-Trajectory-Aware Token Selection","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2602.16169","citing_title":"Discrete Stochastic Localization for Non-autoregressive Generation","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2509.22963","citing_title":"Reinforcement Learning with Discrete Diffusion Policies for Combinatorial Action Spaces","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2511.18801","citing_title":"PartDiffuser: Part-wise 3D Mesh Generation via Discrete Diffusion","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12825","citing_title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15676","citing_title":"Dynamic Chunking for Diffusion Language Models","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2507.12549","citing_title":"The Serial Scaling Hypothesis","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2508.19982","citing_title":"Diffusion Language Models Know the Answer Before Decoding","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","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":5,"is_internal_anchor":true},{"citing_arxiv_id":"2601.20706","citing_title":"NPU Design for Diffusion Language Model Inference","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2604.02340","citing_title":"Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2508.02193","citing_title":"Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2603.12554","citing_title":"Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08564","citing_title":"Attention-Based Sampler for Diffusion Language Models","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13999","citing_title":"Support Before Frequency in Discrete Diffusion","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2603.25661","citing_title":"Fast-dVLA: Accelerating Discrete Diffusion VLA to Real-Time Performance","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12522","citing_title":"Differences in Text Generated by Diffusion and Autoregressive Language Models","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12825","citing_title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2512.15745","citing_title":"LLaDA2.0: Scaling Up Diffusion Language Models to 100B","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2604.02718","citing_title":"Generative Frontiers: Why Evaluation Matters for Diffusion Language Models","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09430","citing_title":"FlashAR: Efficient Post-Training Acceleration for Autoregressive Image Generation","ref_index":1,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC","json":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC.json","graph_json":"https://pith.science/api/pith-number/BTT26H53S5CHOTCCLFFKY2PSXC/graph.json","events_json":"https://pith.science/api/pith-number/BTT26H53S5CHOTCCLFFKY2PSXC/events.json","paper":"https://pith.science/paper/BTT26H53"},"agent_actions":{"view_html":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC","download_json":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC.json","view_paper":"https://pith.science/paper/BTT26H53","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.09573&json=true","fetch_graph":"https://pith.science/api/pith-number/BTT26H53S5CHOTCCLFFKY2PSXC/graph.json","fetch_events":"https://pith.science/api/pith-number/BTT26H53S5CHOTCCLFFKY2PSXC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC/action/storage_attestation","attest_author":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC/action/author_attestation","sign_citation":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC/action/citation_signature","submit_replication":"https://pith.science/pith/BTT26H53S5CHOTCCLFFKY2PSXC/action/replication_record"}},"created_at":"2026-05-17T23:38:52.722484+00:00","updated_at":"2026-05-17T23:38:52.722484+00:00"}