{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GWT7BTE6QQH6QFVV76TWP23RSF","short_pith_number":"pith:GWT7BTE6","schema_version":"1.0","canonical_sha256":"35a7f0cc9e840fe816b5ffa767eb7191736aa50a36f18a9123a3323bad1da014","source":{"kind":"arxiv","id":"2606.22932","version":1},"attestation_state":"computed","paper":{"title":"FORGE: Fused On-Register Gradient Elimination for Memory-Efficient LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aik Beng Ng, Avinash Anand, Bapi Chatterjee, Dikshant Kukreja, Erik Cambria, Kritarth Prasad, Simon See, Timothy Liu, Zhengkui Wang","submitted_at":"2026-06-22T07:08:31Z","abstract_excerpt":"Reverse-mode differentiation computes every weight gradient, writes it to memory, and only then lets the optimizer read it back. This two-phase schedule sets the memory ceiling of modern training: at the seam between the phases, every layer's gradient is live at once. We argue that this materialized gradient is an artifact of how differentiation is staged, not a quantity that learning requires -- and we eliminate it. FORGE folds the optimizer step into the backward pass and applies it one tile at a time, entirely in registers, so each gradient tile is consumed the instant it is produced and ne"},"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":"2606.22932","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-22T07:08:31Z","cross_cats_sorted":[],"title_canon_sha256":"aee27f278770c702d23edb82247120cbaf01af4bea809fa2579332f70b33831b","abstract_canon_sha256":"bce6e2b50307fd0a726d67fe160c3f0f43b68adef2d2b253b394625372758acb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T03:14:04.623115Z","signature_b64":"bdlghSSwiaWo2hObSZgIjxEiueK0G3W/qcG94ix62GbN+hzbAvgUMVieGcafv5xWpBSAsUV1Ubk5r399AALdBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"35a7f0cc9e840fe816b5ffa767eb7191736aa50a36f18a9123a3323bad1da014","last_reissued_at":"2026-06-23T03:14:04.622716Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T03:14:04.622716Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FORGE: Fused On-Register Gradient Elimination for Memory-Efficient LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aik Beng Ng, Avinash Anand, Bapi Chatterjee, Dikshant Kukreja, Erik Cambria, Kritarth Prasad, Simon See, Timothy Liu, Zhengkui Wang","submitted_at":"2026-06-22T07:08:31Z","abstract_excerpt":"Reverse-mode differentiation computes every weight gradient, writes it to memory, and only then lets the optimizer read it back. This two-phase schedule sets the memory ceiling of modern training: at the seam between the phases, every layer's gradient is live at once. We argue that this materialized gradient is an artifact of how differentiation is staged, not a quantity that learning requires -- and we eliminate it. FORGE folds the optimizer step into the backward pass and applies it one tile at a time, entirely in registers, so each gradient tile is consumed the instant it is produced and ne"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.22932","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/2606.22932/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":"2606.22932","created_at":"2026-06-23T03:14:04.622776+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.22932v1","created_at":"2026-06-23T03:14:04.622776+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.22932","created_at":"2026-06-23T03:14:04.622776+00:00"},{"alias_kind":"pith_short_12","alias_value":"GWT7BTE6QQH6","created_at":"2026-06-23T03:14:04.622776+00:00"},{"alias_kind":"pith_short_16","alias_value":"GWT7BTE6QQH6QFVV","created_at":"2026-06-23T03:14:04.622776+00:00"},{"alias_kind":"pith_short_8","alias_value":"GWT7BTE6","created_at":"2026-06-23T03:14:04.622776+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/GWT7BTE6QQH6QFVV76TWP23RSF","json":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF.json","graph_json":"https://pith.science/api/pith-number/GWT7BTE6QQH6QFVV76TWP23RSF/graph.json","events_json":"https://pith.science/api/pith-number/GWT7BTE6QQH6QFVV76TWP23RSF/events.json","paper":"https://pith.science/paper/GWT7BTE6"},"agent_actions":{"view_html":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF","download_json":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF.json","view_paper":"https://pith.science/paper/GWT7BTE6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.22932&json=true","fetch_graph":"https://pith.science/api/pith-number/GWT7BTE6QQH6QFVV76TWP23RSF/graph.json","fetch_events":"https://pith.science/api/pith-number/GWT7BTE6QQH6QFVV76TWP23RSF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF/action/storage_attestation","attest_author":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF/action/author_attestation","sign_citation":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF/action/citation_signature","submit_replication":"https://pith.science/pith/GWT7BTE6QQH6QFVV76TWP23RSF/action/replication_record"}},"created_at":"2026-06-23T03:14:04.622776+00:00","updated_at":"2026-06-23T03:14:04.622776+00:00"}