{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:TTBPLHHJJ5RP3URV6X5V7WGO5I","short_pith_number":"pith:TTBPLHHJ","schema_version":"1.0","canonical_sha256":"9cc2f59ce94f62fdd235f5fb5fd8ceea079c78c82aa58fcf448eb780d97fdc89","source":{"kind":"arxiv","id":"2605.24425","version":1},"attestation_state":"computed","paper":{"title":"Momentum Streams for Optimizer-Inspired Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jiayun Wu, Jingchu Gai, Nai-Chieh Huang","submitted_at":"2026-05-23T06:40:27Z","abstract_excerpt":"The residual update of a pre-norm Transformer layer admits an interpretation as one step of a first-order optimizer acting on a surrogate token energy, wherein the attention and MLP sublayers function as gradient oracles. Based on this observation, we build a family of optimizer-inspired Transformers (triple-momentum, Adam/AdamW, Muon, SOAP) and compare them under matched compute. In our main pretraining experiment, the triple-momentum TMMFormer achieves the lowest validation loss, outperforming the vanilla Transformer and prior architectural variants. A controlled ablation and supporting theo"},"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":"2605.24425","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-23T06:40:27Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a3dac7ae6ea84130e9bff6f3c180515613b1779f6b24e914df56ae3084717c90","abstract_canon_sha256":"c520073f491b032aff2c29b6ed1e55701bf59e576bc8fac46731297aa86e844e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:39.111888Z","signature_b64":"1YScMPWqz1jNTCC6ZkLw9fcZUheHMmrqUQN1tDMU0IYVbRQJJT3mKaFx1AnLZ6xpb8X79VAwgIuIJdbEQvRvAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9cc2f59ce94f62fdd235f5fb5fd8ceea079c78c82aa58fcf448eb780d97fdc89","last_reissued_at":"2026-05-26T01:03:39.111285Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:39.111285Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Momentum Streams for Optimizer-Inspired Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jiayun Wu, Jingchu Gai, Nai-Chieh Huang","submitted_at":"2026-05-23T06:40:27Z","abstract_excerpt":"The residual update of a pre-norm Transformer layer admits an interpretation as one step of a first-order optimizer acting on a surrogate token energy, wherein the attention and MLP sublayers function as gradient oracles. Based on this observation, we build a family of optimizer-inspired Transformers (triple-momentum, Adam/AdamW, Muon, SOAP) and compare them under matched compute. In our main pretraining experiment, the triple-momentum TMMFormer achieves the lowest validation loss, outperforming the vanilla Transformer and prior architectural variants. A controlled ablation and supporting theo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.24425","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/2605.24425/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":"2605.24425","created_at":"2026-05-26T01:03:39.111404+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.24425v1","created_at":"2026-05-26T01:03:39.111404+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.24425","created_at":"2026-05-26T01:03:39.111404+00:00"},{"alias_kind":"pith_short_12","alias_value":"TTBPLHHJJ5RP","created_at":"2026-05-26T01:03:39.111404+00:00"},{"alias_kind":"pith_short_16","alias_value":"TTBPLHHJJ5RP3URV","created_at":"2026-05-26T01:03:39.111404+00:00"},{"alias_kind":"pith_short_8","alias_value":"TTBPLHHJ","created_at":"2026-05-26T01:03:39.111404+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/TTBPLHHJJ5RP3URV6X5V7WGO5I","json":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I.json","graph_json":"https://pith.science/api/pith-number/TTBPLHHJJ5RP3URV6X5V7WGO5I/graph.json","events_json":"https://pith.science/api/pith-number/TTBPLHHJJ5RP3URV6X5V7WGO5I/events.json","paper":"https://pith.science/paper/TTBPLHHJ"},"agent_actions":{"view_html":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I","download_json":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I.json","view_paper":"https://pith.science/paper/TTBPLHHJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.24425&json=true","fetch_graph":"https://pith.science/api/pith-number/TTBPLHHJJ5RP3URV6X5V7WGO5I/graph.json","fetch_events":"https://pith.science/api/pith-number/TTBPLHHJJ5RP3URV6X5V7WGO5I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I/action/storage_attestation","attest_author":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I/action/author_attestation","sign_citation":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I/action/citation_signature","submit_replication":"https://pith.science/pith/TTBPLHHJJ5RP3URV6X5V7WGO5I/action/replication_record"}},"created_at":"2026-05-26T01:03:39.111404+00:00","updated_at":"2026-05-26T01:03:39.111404+00:00"}