{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3GYUN2URXR4SPLVCCPMWVP2NLY","short_pith_number":"pith:3GYUN2UR","schema_version":"1.0","canonical_sha256":"d9b146ea91bc7927aea213d96abf4d5e3e4faf439bca2cfbad570922b03a5b60","source":{"kind":"arxiv","id":"2508.18756","version":1},"attestation_state":"computed","paper":{"title":"UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Banggu Wu, Defa Zhu, Hongzhi Huang, Qiyang Min, Ran Guo, Siyan Chen, Siyuan Qiao, Xun Zhou, Yu Bao, Yutao Zeng, Zihao Huang","submitted_at":"2025-08-26T07:33:11Z","abstract_excerpt":"While Mixture of Experts (MoE) models achieve remarkable efficiency by activating only subsets of parameters, they suffer from high memory access costs during inference. Memory-layer architectures offer an appealing alternative with very few memory access, but previous attempts like UltraMem have only matched the performance of 2-expert MoE models, falling significantly short of state-of-the-art 8-expert configurations. We present UltraMemV2, a redesigned memory-layer architecture that closes this performance gap. Our approach introduces five key improvements: integrating memory layers into ev"},"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":"2508.18756","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-08-26T07:33:11Z","cross_cats_sorted":[],"title_canon_sha256":"c1a9aa2c565c61950e91088223854ebe008931488f47e65d11972ba7f28ee073","abstract_canon_sha256":"a2f38cc38a2d019bd261b3101b0273068ed9e8cf7cfcdf51e61435f55899f190"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:59:30.290028Z","signature_b64":"jZJSVegK+KXo9QK2J+PJL/DK4Y8XyvzEqHcJGVxZ2PSBUqYqzXsnBYNmEHsVedaCwyvdnuKEt7fHYDMQ5KHbAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d9b146ea91bc7927aea213d96abf4d5e3e4faf439bca2cfbad570922b03a5b60","last_reissued_at":"2026-07-05T11:59:30.289607Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:59:30.289607Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Banggu Wu, Defa Zhu, Hongzhi Huang, Qiyang Min, Ran Guo, Siyan Chen, Siyuan Qiao, Xun Zhou, Yu Bao, Yutao Zeng, Zihao Huang","submitted_at":"2025-08-26T07:33:11Z","abstract_excerpt":"While Mixture of Experts (MoE) models achieve remarkable efficiency by activating only subsets of parameters, they suffer from high memory access costs during inference. Memory-layer architectures offer an appealing alternative with very few memory access, but previous attempts like UltraMem have only matched the performance of 2-expert MoE models, falling significantly short of state-of-the-art 8-expert configurations. We present UltraMemV2, a redesigned memory-layer architecture that closes this performance gap. Our approach introduces five key improvements: integrating memory layers into ev"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.18756","kind":"arxiv","version":1},"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/2508.18756/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":"2508.18756","created_at":"2026-07-05T11:59:30.289665+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.18756v1","created_at":"2026-07-05T11:59:30.289665+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.18756","created_at":"2026-07-05T11:59:30.289665+00:00"},{"alias_kind":"pith_short_12","alias_value":"3GYUN2URXR4S","created_at":"2026-07-05T11:59:30.289665+00:00"},{"alias_kind":"pith_short_16","alias_value":"3GYUN2URXR4SPLVC","created_at":"2026-07-05T11:59:30.289665+00:00"},{"alias_kind":"pith_short_8","alias_value":"3GYUN2UR","created_at":"2026-07-05T11:59:30.289665+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09888","citing_title":"SinkRec: Mitigating Semantic State Sink in Long Sequence Recommendation with Memory-Conditioned Gated Delta Networks","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2512.13751","citing_title":"MIDUS: Memory-Infused Depth Up-Scaling","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY","json":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY.json","graph_json":"https://pith.science/api/pith-number/3GYUN2URXR4SPLVCCPMWVP2NLY/graph.json","events_json":"https://pith.science/api/pith-number/3GYUN2URXR4SPLVCCPMWVP2NLY/events.json","paper":"https://pith.science/paper/3GYUN2UR"},"agent_actions":{"view_html":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY","download_json":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY.json","view_paper":"https://pith.science/paper/3GYUN2UR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.18756&json=true","fetch_graph":"https://pith.science/api/pith-number/3GYUN2URXR4SPLVCCPMWVP2NLY/graph.json","fetch_events":"https://pith.science/api/pith-number/3GYUN2URXR4SPLVCCPMWVP2NLY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY/action/storage_attestation","attest_author":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY/action/author_attestation","sign_citation":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY/action/citation_signature","submit_replication":"https://pith.science/pith/3GYUN2URXR4SPLVCCPMWVP2NLY/action/replication_record"}},"created_at":"2026-07-05T11:59:30.289665+00:00","updated_at":"2026-07-05T11:59:30.289665+00:00"}