{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IWV33LYCYZFR66P3AX3H6VBJM3","short_pith_number":"pith:IWV33LYC","schema_version":"1.0","canonical_sha256":"45abbdaf02c64b1f79fb05f67f542966e5d8f7760999c4ac8f513bb992e20372","source":{"kind":"arxiv","id":"2601.23278","version":2},"attestation_state":"computed","paper":{"title":"FOCUS: DLLMs Know How to Tame Their Compute Bound","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AR","cs.CL"],"primary_cat":"cs.LG","authors_text":"An Zhong, Hong Xu, Kaihua Liang, Marco Canini, Xin Tan","submitted_at":"2026-01-30T18:52:06Z","abstract_excerpt":"Diffusion Large Language Models (DLLMs) offer a compelling alternative to Auto-Regressive models, but their deployment is constrained by high decoding cost. In this work, we identify a key inefficiency in DLLM decoding: while computation is parallelized over token blocks, only a small subset of tokens is decodable at each diffusion step, causing most compute to be wasted on non-decodable tokens. We further observe a strong correlation between attention-derived token importance and token-wise decoding probability. Based on this insight, we propose FOCUS, an inference system designed for DLLMs. "},"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":"2601.23278","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-30T18:52:06Z","cross_cats_sorted":["cs.AR","cs.CL"],"title_canon_sha256":"65181a7a96d9e4cc20d75a29cbff711a6d32fdfe72b99a6d0768f4f664f21d32","abstract_canon_sha256":"0bf16096d91afa0b951299502f26c36312e97f675317ed9dd272fb0efaaa3f92"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-11T01:10:32.193977Z","signature_b64":"tddBEAPugxkeqL0D6XUpuVBecl0Q9MpZw/z6pnJ43gKeZ1t9/kCfUaobNFMFKHUx+3xCS6nijBfx2X2G6nCJBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"45abbdaf02c64b1f79fb05f67f542966e5d8f7760999c4ac8f513bb992e20372","last_reissued_at":"2026-06-11T01:10:32.192929Z","signature_status":"signed_v1","first_computed_at":"2026-06-11T01:10:32.192929Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FOCUS: DLLMs Know How to Tame Their Compute Bound","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AR","cs.CL"],"primary_cat":"cs.LG","authors_text":"An Zhong, Hong Xu, Kaihua Liang, Marco Canini, Xin Tan","submitted_at":"2026-01-30T18:52:06Z","abstract_excerpt":"Diffusion Large Language Models (DLLMs) offer a compelling alternative to Auto-Regressive models, but their deployment is constrained by high decoding cost. In this work, we identify a key inefficiency in DLLM decoding: while computation is parallelized over token blocks, only a small subset of tokens is decodable at each diffusion step, causing most compute to be wasted on non-decodable tokens. We further observe a strong correlation between attention-derived token importance and token-wise decoding probability. Based on this insight, we propose FOCUS, an inference system designed for DLLMs. "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.23278","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/2601.23278/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":"2601.23278","created_at":"2026-06-11T01:10:32.193078+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.23278v2","created_at":"2026-06-11T01:10:32.193078+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.23278","created_at":"2026-06-11T01:10:32.193078+00:00"},{"alias_kind":"pith_short_12","alias_value":"IWV33LYCYZFR","created_at":"2026-06-11T01:10:32.193078+00:00"},{"alias_kind":"pith_short_16","alias_value":"IWV33LYCYZFR66P3","created_at":"2026-06-11T01:10:32.193078+00:00"},{"alias_kind":"pith_short_8","alias_value":"IWV33LYC","created_at":"2026-06-11T01:10:32.193078+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3","json":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3.json","graph_json":"https://pith.science/api/pith-number/IWV33LYCYZFR66P3AX3H6VBJM3/graph.json","events_json":"https://pith.science/api/pith-number/IWV33LYCYZFR66P3AX3H6VBJM3/events.json","paper":"https://pith.science/paper/IWV33LYC"},"agent_actions":{"view_html":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3","download_json":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3.json","view_paper":"https://pith.science/paper/IWV33LYC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.23278&json=true","fetch_graph":"https://pith.science/api/pith-number/IWV33LYCYZFR66P3AX3H6VBJM3/graph.json","fetch_events":"https://pith.science/api/pith-number/IWV33LYCYZFR66P3AX3H6VBJM3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3/action/storage_attestation","attest_author":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3/action/author_attestation","sign_citation":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3/action/citation_signature","submit_replication":"https://pith.science/pith/IWV33LYCYZFR66P3AX3H6VBJM3/action/replication_record"}},"created_at":"2026-06-11T01:10:32.193078+00:00","updated_at":"2026-06-11T01:10:32.193078+00:00"}