{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NTWZA4UIH5XFDMAU3BV2QNB3OI","short_pith_number":"pith:NTWZA4UI","schema_version":"1.0","canonical_sha256":"6ced9072883f6e51b014d86ba8343b721440c455dacebe62c1984fdc7475b308","source":{"kind":"arxiv","id":"2605.12825","version":1},"attestation_state":"computed","paper":{"title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Orthrus adds a lightweight diffusion view to frozen LLMs so they can generate tokens in parallel while matching standard autoregressive output exactly.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chaitra Hegde, Chien Van Nguyen, Franck Dernoncourt, Ryan A. Rossi, Thien Huu Nguyen, Van Cuong Pham","submitted_at":"2026-05-12T23:47:35Z","abstract_excerpt":"We introduce Orthrus, a simple and efficient dual-architecture framework that unifies the exact generation fidelity of autoregressive Large Language Models (LLMs) with the high-speed parallel token generation of diffusion models. The sequential nature of standard autoregressive decoding represents a fundamental bottleneck for high-throughput inference. While diffusion language models attempt to break this barrier via parallel generation, they suffer from significant performance degradation, high training costs, and a lack of rigorous convergence guarantees. Orthrus resolves this dichotomy nati"},"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":false},"canonical_record":{"source":{"id":"2605.12825","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:47:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"05da86b68e5d5d3b1f60a664c9b3d780981d51cb2252d88d47fccda2f5af9d48","abstract_canon_sha256":"a55b2eafd4e5b857c91d8247ac615ad8f7469d957024becaad6dc1365e4203a2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:12.166174Z","signature_b64":"MSH8qkEyfuafl2j2MJ+yuuPgc7pqrtOBxgJmNq1xlhPlUkotgOETHli9XvnyJQ18G3LnKyClzQKq2tE5TA4hCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6ced9072883f6e51b014d86ba8343b721440c455dacebe62c1984fdc7475b308","last_reissued_at":"2026-05-18T03:09:12.165558Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:12.165558Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Orthrus adds a lightweight diffusion view to frozen LLMs so they can generate tokens in parallel while matching standard autoregressive output exactly.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chaitra Hegde, Chien Van Nguyen, Franck Dernoncourt, Ryan A. Rossi, Thien Huu Nguyen, Van Cuong Pham","submitted_at":"2026-05-12T23:47:35Z","abstract_excerpt":"We introduce Orthrus, a simple and efficient dual-architecture framework that unifies the exact generation fidelity of autoregressive Large Language Models (LLMs) with the high-speed parallel token generation of diffusion models. The sequential nature of standard autoregressive decoding represents a fundamental bottleneck for high-throughput inference. While diffusion language models attempt to break this barrier via parallel generation, they suffer from significant performance degradation, high training costs, and a lack of rigorous convergence guarantees. Orthrus resolves this dichotomy nati"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By employing an exact consensus mechanism between the two views, Orthrus guarantees lossless inference, delivering up to a 7.8x speedup with only an O(1) memory cache overhead and minimal parameter additions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The exact consensus mechanism between the autoregressive and diffusion views will always produce identical output to standard autoregressive decoding without introducing any quality degradation or requiring additional training data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Orthrus unifies autoregressive and diffusion views on a shared KV cache to deliver lossless parallel token generation with up to 7.8x speedup and O(1) memory overhead.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Orthrus adds a lightweight diffusion view to frozen LLMs so they can generate tokens in parallel while matching standard autoregressive output exactly.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d91d4f3a7de07a738ebb93f26dab1c45d9399a3d7ee92209e9882dda63fe4fa6"},"source":{"id":"2605.12825","kind":"arxiv","version":1},"verdict":{"id":"73366870-8aa7-4fae-83bd-b4a11c61d335","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:44:26.418005Z","strongest_claim":"By employing an exact consensus mechanism between the two views, Orthrus guarantees lossless inference, delivering up to a 7.8x speedup with only an O(1) memory cache overhead and minimal parameter additions.","one_line_summary":"Orthrus unifies autoregressive and diffusion views on a shared KV cache to deliver lossless parallel token generation with up to 7.8x speedup and O(1) memory overhead.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The exact consensus mechanism between the autoregressive and diffusion views will always produce identical output to standard autoregressive decoding without introducing any quality degradation or requiring additional training data.","pith_extraction_headline":"Orthrus adds a lightweight diffusion view to frozen LLMs so they can generate tokens in parallel while matching standard autoregressive output exactly."},"references":{"count":25,"sample":[{"doi":"","year":null,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","ref_index":1,"cited_arxiv_id":"2303.08774","is_internal_anchor":true},{"doi":"","year":null,"title":"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models","work_id":"b34ab928-6ffb-4028-b13c-395a8924d76b","ref_index":2,"cited_arxiv_id":"2503.09573","is_internal_anchor":true},{"doi":"","year":2026,"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","ref_index":3,"cited_arxiv_id":"2108.07732","is_internal_anchor":true},{"doi":"","year":1901,"title":"Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901","work_id":"84780adb-76cf-44ac-a8b7-e24d4fa5c592","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Zhuoming Chen, Avner May, Ruslan Svirschevski, Yuhsun Huang, Max Ryabinin, Zhihao Jia, and Beidi Chen","work_id":"860ff726-6feb-4917-a25f-95159b43dd49","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":25,"snapshot_sha256":"ec10b49f3834071957bf692f99aab5386713c79f586afa0254c8f0a6d037e9ff","internal_anchors":14},"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":"2605.12825","created_at":"2026-05-18T03:09:12.165648+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.12825v1","created_at":"2026-05-18T03:09:12.165648+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12825","created_at":"2026-05-18T03:09:12.165648+00:00"},{"alias_kind":"pith_short_12","alias_value":"NTWZA4UIH5XF","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"NTWZA4UIH5XFDMAU","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"NTWZA4UI","created_at":"2026-05-18T12:33:37.589309+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/NTWZA4UIH5XFDMAU3BV2QNB3OI","json":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI.json","graph_json":"https://pith.science/api/pith-number/NTWZA4UIH5XFDMAU3BV2QNB3OI/graph.json","events_json":"https://pith.science/api/pith-number/NTWZA4UIH5XFDMAU3BV2QNB3OI/events.json","paper":"https://pith.science/paper/NTWZA4UI"},"agent_actions":{"view_html":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI","download_json":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI.json","view_paper":"https://pith.science/paper/NTWZA4UI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.12825&json=true","fetch_graph":"https://pith.science/api/pith-number/NTWZA4UIH5XFDMAU3BV2QNB3OI/graph.json","fetch_events":"https://pith.science/api/pith-number/NTWZA4UIH5XFDMAU3BV2QNB3OI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/action/storage_attestation","attest_author":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/action/author_attestation","sign_citation":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/action/citation_signature","submit_replication":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/action/replication_record"}},"created_at":"2026-05-18T03:09:12.165648+00:00","updated_at":"2026-05-18T03:09:12.165648+00:00"}