{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:NTWZA4UIH5XFDMAU3BV2QNB3OI","short_pith_number":"pith:NTWZA4UI","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"},"canonical_sha256":"6ced9072883f6e51b014d86ba8343b721440c455dacebe62c1984fdc7475b308","source":{"kind":"arxiv","id":"2605.12825","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12825","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12825v1","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12825","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"pith_short_12","alias_value":"NTWZA4UIH5XF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"NTWZA4UIH5XFDMAU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"NTWZA4UI","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:NTWZA4UIH5XFDMAU3BV2QNB3OI","target":"record","payload":{"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"},"canonical_sha256":"6ced9072883f6e51b014d86ba8343b721440c455dacebe62c1984fdc7475b308","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"},"source_kind":"arxiv","source_id":"2605.12825","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"r+AW+Xts7xicVAOWadHRBCe4GDt6CLnfqNLwTcnewYaKV/eKmmTAMIvKXn3UJeV0iZ5w47EWR4eRjzCqw9w1AQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T23:00:21.814415Z"},"content_sha256":"dc8c2ff3cc39135d6056e2370c5e5dde07de7be4101d9e936120eff863ae7e85","schema_version":"1.0","event_id":"sha256:dc8c2ff3cc39135d6056e2370c5e5dde07de7be4101d9e936120eff863ae7e85"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:NTWZA4UIH5XFDMAU3BV2QNB3OI","target":"graph","payload":{"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"},"verdict_id":"73366870-8aa7-4fae-83bd-b4a11c61d335"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ImoL40Kw2YlXlLex/oNGxIbV2dJKQmJWCEjwhL1zXrS0Ryw9Xosq5H+u0rDkYrNF4eOyv6usHBc+BniRouftAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T23:00:21.815107Z"},"content_sha256":"fb609fa653421c54d41572fa966c9f9a70b498c8266d249e6a24279f075dc49a","schema_version":"1.0","event_id":"sha256:fb609fa653421c54d41572fa966c9f9a70b498c8266d249e6a24279f075dc49a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/bundle.json","state_url":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-28T23:00:21Z","links":{"resolver":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI","bundle":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/bundle.json","state":"https://pith.science/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/NTWZA4UIH5XFDMAU3BV2QNB3OI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:NTWZA4UIH5XFDMAU3BV2QNB3OI","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"a55b2eafd4e5b857c91d8247ac615ad8f7469d957024becaad6dc1365e4203a2","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:47:35Z","title_canon_sha256":"05da86b68e5d5d3b1f60a664c9b3d780981d51cb2252d88d47fccda2f5af9d48"},"schema_version":"1.0","source":{"id":"2605.12825","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12825","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12825v1","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12825","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"pith_short_12","alias_value":"NTWZA4UIH5XF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"NTWZA4UIH5XFDMAU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"NTWZA4UI","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:fb609fa653421c54d41572fa966c9f9a70b498c8266d249e6a24279f075dc49a","target":"graph","created_at":"2026-05-18T03:09:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Orthrus adds a lightweight diffusion view to frozen LLMs so they can generate tokens in parallel while matching standard autoregressive output exactly."}],"snapshot_sha256":"d91d4f3a7de07a738ebb93f26dab1c45d9399a3d7ee92209e9882dda63fe4fa6"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"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","authors_text":"Chaitra Hegde, Chien Van Nguyen, Franck Dernoncourt, Ryan A. Rossi, Thien Huu Nguyen, Van Cuong Pham","cross_cats":["cs.AI"],"headline":"Orthrus adds a lightweight diffusion view to frozen LLMs so they can generate tokens in parallel while matching standard autoregressive output exactly.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:47:35Z","title":"Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion"},"references":{"count":25,"internal_anchors":14,"resolved_work":25,"sample":[{"cited_arxiv_id":"2303.08774","doi":"","is_internal_anchor":true,"ref_index":1,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","year":null},{"cited_arxiv_id":"2503.09573","doi":"","is_internal_anchor":true,"ref_index":2,"title":"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models","work_id":"b34ab928-6ffb-4028-b13c-395a8924d76b","year":null},{"cited_arxiv_id":"2108.07732","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","year":2026},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901","work_id":"84780adb-76cf-44ac-a8b7-e24d4fa5c592","year":1901},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Zhuoming Chen, Avner May, Ruslan Svirschevski, Yuhsun Huang, Max Ryabinin, Zhihao Jia, and Beidi Chen","work_id":"860ff726-6feb-4917-a25f-95159b43dd49","year":null}],"snapshot_sha256":"ec10b49f3834071957bf692f99aab5386713c79f586afa0254c8f0a6d037e9ff"},"source":{"id":"2605.12825","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:44:26.418005Z","id":"73366870-8aa7-4fae-83bd-b4a11c61d335","model_set":{"reader":"grok-4.3"},"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","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.","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.","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."}},"verdict_id":"73366870-8aa7-4fae-83bd-b4a11c61d335"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:dc8c2ff3cc39135d6056e2370c5e5dde07de7be4101d9e936120eff863ae7e85","target":"record","created_at":"2026-05-18T03:09:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"a55b2eafd4e5b857c91d8247ac615ad8f7469d957024becaad6dc1365e4203a2","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:47:35Z","title_canon_sha256":"05da86b68e5d5d3b1f60a664c9b3d780981d51cb2252d88d47fccda2f5af9d48"},"schema_version":"1.0","source":{"id":"2605.12825","kind":"arxiv","version":1}},"canonical_sha256":"6ced9072883f6e51b014d86ba8343b721440c455dacebe62c1984fdc7475b308","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6ced9072883f6e51b014d86ba8343b721440c455dacebe62c1984fdc7475b308","first_computed_at":"2026-05-18T03:09:12.165558Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:12.165558Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"MSH8qkEyfuafl2j2MJ+yuuPgc7pqrtOBxgJmNq1xlhPlUkotgOETHli9XvnyJQ18G3LnKyClzQKq2tE5TA4hCQ==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:12.166174Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12825","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:dc8c2ff3cc39135d6056e2370c5e5dde07de7be4101d9e936120eff863ae7e85","sha256:fb609fa653421c54d41572fa966c9f9a70b498c8266d249e6a24279f075dc49a"],"state_sha256":"6be877b9fcae46fd1d88d21354beaf31dccab0392c67b442461c7ccd5f66adce"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/dvyTg/yHHoYJEE3e3o8CZHShWF+hhpe1A42w3/5sphuiYS1pXSx4PGqA7jisMzTfoiyRl9MmStqIXswZJVYDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T23:00:21.817805Z","bundle_sha256":"f7784d58f9c236228973881d9bb8aa4bf9263f5efe90f6dfb82c5a10f936db94"}}