{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:K65KSBEJ66G4FSPYQIYEDZD52A","short_pith_number":"pith:K65KSBEJ","schema_version":"1.0","canonical_sha256":"57baa90489f78dc2c9f8823041e47dd006e98d40698ec7d2184f90e4107eac2e","source":{"kind":"arxiv","id":"2405.07719","version":5},"attestation_state":"computed","paper":{"title":"USP: A Unified Sequence Parallelism Approach for Long Context Generative AI","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jiarui Fang, Shangchun Zhao","submitted_at":"2024-05-13T13:08:02Z","abstract_excerpt":"Sequence parallelism (SP), which divides the sequence dimension of input tensors across multiple computational devices, is becoming key to unlocking the long-context capabilities of generative AI models. This paper investigates the state-of-the-art SP approaches, i.e. DeepSpeed-Ulysses and Ring-Attention, and proposes a unified SP approach, which is more robust to transformer model architectures and network hardware topology. This paper compares the communication and memory cost of SP and existing parallelism, including data/tensor/zero/pipeline parallelism, and discusses the best practices fo"},"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":"2405.07719","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-13T13:08:02Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"3bf64ad36350737a4cb4973d3df7e26ad94af2d570f655552672ee72d8ed48e6","abstract_canon_sha256":"a278bdd1841a139b005d51d962ec754deee7e87377edccd3b1312e400349f5eb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:38:59.977129Z","signature_b64":"rSwgIyU8hkM7UWj2en3pugOLFGcHH5yEGXwTNPvHXg49ydeJI41RI+IroADUsCxqgmSDDvqut3lbTBQvbVDFBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"57baa90489f78dc2c9f8823041e47dd006e98d40698ec7d2184f90e4107eac2e","last_reissued_at":"2026-07-05T08:38:59.976636Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:38:59.976636Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"USP: A Unified Sequence Parallelism Approach for Long Context Generative AI","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jiarui Fang, Shangchun Zhao","submitted_at":"2024-05-13T13:08:02Z","abstract_excerpt":"Sequence parallelism (SP), which divides the sequence dimension of input tensors across multiple computational devices, is becoming key to unlocking the long-context capabilities of generative AI models. This paper investigates the state-of-the-art SP approaches, i.e. DeepSpeed-Ulysses and Ring-Attention, and proposes a unified SP approach, which is more robust to transformer model architectures and network hardware topology. This paper compares the communication and memory cost of SP and existing parallelism, including data/tensor/zero/pipeline parallelism, and discusses the best practices fo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.07719","kind":"arxiv","version":5},"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/2405.07719/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":"2405.07719","created_at":"2026-07-05T08:38:59.976694+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.07719v5","created_at":"2026-07-05T08:38:59.976694+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.07719","created_at":"2026-07-05T08:38:59.976694+00:00"},{"alias_kind":"pith_short_12","alias_value":"K65KSBEJ66G4","created_at":"2026-07-05T08:38:59.976694+00:00"},{"alias_kind":"pith_short_16","alias_value":"K65KSBEJ66G4FSPY","created_at":"2026-07-05T08:38:59.976694+00:00"},{"alias_kind":"pith_short_8","alias_value":"K65KSBEJ","created_at":"2026-07-05T08:38:59.976694+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06216","citing_title":"MoWorld: A Flash World Model","ref_index":56,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12195","citing_title":"InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoning","ref_index":256,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12688","citing_title":"M*: A Modular, Extensible, Serving System for Multimodal Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28691","citing_title":"OSP-Next: Efficient High-Quality Video Generation with Sparse Sequence Parallelism, HiF8 Quantization, and Reinforcement Learning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2405.14430","citing_title":"PipeFusion: Patch-level Pipeline Parallelism for Diffusion Transformers Inference","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2503.20314","citing_title":"Wan: Open and Advanced Large-Scale Video Generative Models","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18830","citing_title":"MTraining: Distributed Dynamic Sparse Attention for Efficient Ultra-Long Context Training","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18739","citing_title":"LongLive-2.0: An NVFP4 Parallel Infrastructure for Long Video Generation","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17923","citing_title":"AdaptiveLoad: Towards Efficient Video Diffusion Transformer Training","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2408.10188","citing_title":"LongVILA: Scaling Long-Context Visual Language Models for Long Videos","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2501.12386","citing_title":"InternVideo2.5: Empowering Video MLLMs with Long and Rich Context Modeling","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2505.23884","citing_title":"Test-Time Training Done Right","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2502.13189","citing_title":"MoBA: Mixture of Block Attention for Long-Context LLMs","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2505.13211","citing_title":"MAGI-1: Autoregressive Video Generation at Scale","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07569","citing_title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06168","citing_title":"Action Images: End-to-End Policy Learning via Multiview Video Generation","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05182","citing_title":"LSRM: High-Fidelity Object-Centric Reconstruction via Scaled Context Windows","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14561","citing_title":"CoCoDiff: Optimizing Collective Communications for Distributed Diffusion Transformer Inference Under Ulysses Sequence Parallelism","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02641","citing_title":"Mamoda2.5: Enhancing Unified Multimodal Model with DiT-MoE","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02641","citing_title":"Mamoda2.5: Enhancing Unified Multimodal Model with DiT-MoE","ref_index":44,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A","json":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A.json","graph_json":"https://pith.science/api/pith-number/K65KSBEJ66G4FSPYQIYEDZD52A/graph.json","events_json":"https://pith.science/api/pith-number/K65KSBEJ66G4FSPYQIYEDZD52A/events.json","paper":"https://pith.science/paper/K65KSBEJ"},"agent_actions":{"view_html":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A","download_json":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A.json","view_paper":"https://pith.science/paper/K65KSBEJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.07719&json=true","fetch_graph":"https://pith.science/api/pith-number/K65KSBEJ66G4FSPYQIYEDZD52A/graph.json","fetch_events":"https://pith.science/api/pith-number/K65KSBEJ66G4FSPYQIYEDZD52A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A/action/storage_attestation","attest_author":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A/action/author_attestation","sign_citation":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A/action/citation_signature","submit_replication":"https://pith.science/pith/K65KSBEJ66G4FSPYQIYEDZD52A/action/replication_record"}},"created_at":"2026-07-05T08:38:59.976694+00:00","updated_at":"2026-07-05T08:38:59.976694+00:00"}