{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:KQGQ3HUFVD5QJXTDZURY2OGC6I","short_pith_number":"pith:KQGQ3HUF","canonical_record":{"source":{"id":"2604.07663","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-09T00:07:38Z","cross_cats_sorted":[],"title_canon_sha256":"aa16abe577c9cc3b4f2638308385ac68083532f098b0c3f4f4a2f80d4b295b19","abstract_canon_sha256":"5a8f77f162041ef977b19efbf3b0e41f621029f50f0f7f23836c3f9a532a3e23"},"schema_version":"1.0"},"canonical_sha256":"540d0d9e85a8fb04de63cd238d38c2f2182086c8d712ff6ee20cb40ffaa9e74d","source":{"kind":"arxiv","id":"2604.07663","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.07663","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"arxiv_version","alias_value":"2604.07663v2","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.07663","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"pith_short_12","alias_value":"KQGQ3HUFVD5Q","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"pith_short_16","alias_value":"KQGQ3HUFVD5QJXTD","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"pith_short_8","alias_value":"KQGQ3HUF","created_at":"2026-06-26T01:15:51Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:KQGQ3HUFVD5QJXTDZURY2OGC6I","target":"record","payload":{"canonical_record":{"source":{"id":"2604.07663","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-09T00:07:38Z","cross_cats_sorted":[],"title_canon_sha256":"aa16abe577c9cc3b4f2638308385ac68083532f098b0c3f4f4a2f80d4b295b19","abstract_canon_sha256":"5a8f77f162041ef977b19efbf3b0e41f621029f50f0f7f23836c3f9a532a3e23"},"schema_version":"1.0"},"canonical_sha256":"540d0d9e85a8fb04de63cd238d38c2f2182086c8d712ff6ee20cb40ffaa9e74d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T01:15:51.382405Z","signature_b64":"YEFqOUzepbftRQ3FGXbPtW4Ax9j3Nh+O8QJweaC/e5JHlrIBeDnpIdAygtVWg7LTAAFaBhh3Y+EKldUvNG+uCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"540d0d9e85a8fb04de63cd238d38c2f2182086c8d712ff6ee20cb40ffaa9e74d","last_reissued_at":"2026-06-26T01:15:51.382011Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T01:15:51.382011Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.07663","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-26T01:15:51Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Sy6FRA3NXYMiUA9A58n3zicQ1IjqN15w8AJdYo+HsrgokJTxGpX8EwE57K6BjTPcHSx9a1KjTYAzrA/HMJDJAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-26T04:38:55.086436Z"},"content_sha256":"e30b2641b931d75633f42e6aa241a18f470145e81461f469e8ebd6ef46cd0023","schema_version":"1.0","event_id":"sha256:e30b2641b931d75633f42e6aa241a18f470145e81461f469e8ebd6ef46cd0023"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:KQGQ3HUFVD5QJXTDZURY2OGC6I","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SAGE: Sign-Adaptive Gradient for Memory-Efficient LLM Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SAGE replaces AdamW in hybrid LLM optimizers by adding a bounded per-dimension scale that tames embedding-layer gradients, cutting memory while raising perplexity performance.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hyun-Tae Kim, Wooin Lee","submitted_at":"2026-04-09T00:07:38Z","abstract_excerpt":"The AdamW optimizer, while standard for LLM pretraining, is a critical memory bottleneck, consuming optimizer states equivalent to twice the model's size. Although light-state optimizers like SinkGD attempt to address this issue, we identify the embedding layer dilemma: these methods fail to handle the sparse, high-variance gradients inherent to embeddings, forcing a hybrid design that reverts to AdamW and partially negates the memory gains. We propose SAGE (Sign Adaptive GradiEnt), a novel optimizer that resolves this dilemma by replacing AdamW in this hybrid structure. SAGE combines a Lion-s"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On Llama models up to 1.3B parameters, our SAGE-based hybrid achieves new state-of-the-art perplexity, outperforming all baselines, including SinkGD hybrid, while significantly reducing optimizer state memory.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the high-variance, sparse gradients in the embedding layer can be sufficiently controlled by a single per-dimension scale provably bounded by 1.0 without reintroducing the instability that forced prior methods back to AdamW.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SAGE replaces AdamW in memory-efficient LLM hybrids with a Lion-style sign update plus a provably bounded O(d) adaptive scale, delivering SOTA perplexity on Llama-1.3B while cutting optimizer-state memory.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SAGE replaces AdamW in hybrid LLM optimizers by adding a bounded per-dimension scale that tames embedding-layer gradients, cutting memory while raising perplexity performance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9959ef4638bbf38b963afe1a2259eda47df1e90fb591bc7e06d3b07a3a633033"},"source":{"id":"2604.07663","kind":"arxiv","version":2},"verdict":{"id":"b0946ce6-2077-4e74-877c-38f3fb057507","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T18:33:19.243230Z","strongest_claim":"On Llama models up to 1.3B parameters, our SAGE-based hybrid achieves new state-of-the-art perplexity, outperforming all baselines, including SinkGD hybrid, while significantly reducing optimizer state memory.","one_line_summary":"SAGE replaces AdamW in memory-efficient LLM hybrids with a Lion-style sign update plus a provably bounded O(d) adaptive scale, delivering SOTA perplexity on Llama-1.3B while cutting optimizer-state memory.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the high-variance, sparse gradients in the embedding layer can be sufficiently controlled by a single per-dimension scale provably bounded by 1.0 without reintroducing the instability that forced prior methods back to AdamW.","pith_extraction_headline":"SAGE replaces AdamW in hybrid LLM optimizers by adding a bounded per-dimension scale that tames embedding-layer gradients, cutting memory while raising perplexity performance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.07663/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":3,"sample":[{"doi":"","year":2016,"title":"Layer Normalization","work_id":"20a2d720-0046-4c7c-bcd6-327ec8143f69","ref_index":1,"cited_arxiv_id":"1607.06450","is_internal_anchor":true},{"doi":"","year":2024,"title":"DAtts L ⊙ 1√ h softmax′ gAtt s L√ h !# Y K L , DY K L =","work_id":"fe712d95-0ed1-4d1a-94aa-b36a44a56919","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2048,"title":"Opt.\" refers to optimizer state memory, while","work_id":"68f80759-6f11-4808-8e11-3247f4a03dd9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":3,"snapshot_sha256":"2577a907f5d50e501c1dd2a2c777f4dce9b45a3f6abf1b802efbb7184d046c09","internal_anchors":1},"formal_canon":{"evidence_count":1,"snapshot_sha256":"d2e5cd0875624164da59059db0f1aca195c89aacecf5a84d78bc5af301a8de37"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"b0946ce6-2077-4e74-877c-38f3fb057507"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-26T01:15:51Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"KV9SDIAsmnLzoCAnJwOGbC+wlhRfUBdrcLRrbbDlvl3fmax2rZATIyXYVErlABdvgQ6EPP66RD6OF/O9IHzWBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-26T04:38:55.087047Z"},"content_sha256":"2264cac6a464e4a589ed77b48b6c92394f92a8e8332ba9f7e52c50625a5f31f2","schema_version":"1.0","event_id":"sha256:2264cac6a464e4a589ed77b48b6c92394f92a8e8332ba9f7e52c50625a5f31f2"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I/bundle.json","state_url":"https://pith.science/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I/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":"2026-06-26T04:38:55Z","links":{"resolver":"https://pith.science/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I","bundle":"https://pith.science/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I/bundle.json","state":"https://pith.science/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I/state.json","well_known_bundle":"https://pith.science/.well-known/pith/KQGQ3HUFVD5QJXTDZURY2OGC6I/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:KQGQ3HUFVD5QJXTDZURY2OGC6I","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":"5a8f77f162041ef977b19efbf3b0e41f621029f50f0f7f23836c3f9a532a3e23","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-09T00:07:38Z","title_canon_sha256":"aa16abe577c9cc3b4f2638308385ac68083532f098b0c3f4f4a2f80d4b295b19"},"schema_version":"1.0","source":{"id":"2604.07663","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.07663","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"arxiv_version","alias_value":"2604.07663v2","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.07663","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"pith_short_12","alias_value":"KQGQ3HUFVD5Q","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"pith_short_16","alias_value":"KQGQ3HUFVD5QJXTD","created_at":"2026-06-26T01:15:51Z"},{"alias_kind":"pith_short_8","alias_value":"KQGQ3HUF","created_at":"2026-06-26T01:15:51Z"}],"graph_snapshots":[{"event_id":"sha256:2264cac6a464e4a589ed77b48b6c92394f92a8e8332ba9f7e52c50625a5f31f2","target":"graph","created_at":"2026-06-26T01:15:51Z","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":"On Llama models up to 1.3B parameters, our SAGE-based hybrid achieves new state-of-the-art perplexity, outperforming all baselines, including SinkGD hybrid, while significantly reducing optimizer state memory."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the high-variance, sparse gradients in the embedding layer can be sufficiently controlled by a single per-dimension scale provably bounded by 1.0 without reintroducing the instability that forced prior methods back to AdamW."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SAGE replaces AdamW in memory-efficient LLM hybrids with a Lion-style sign update plus a provably bounded O(d) adaptive scale, delivering SOTA perplexity on Llama-1.3B while cutting optimizer-state memory."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"SAGE replaces AdamW in hybrid LLM optimizers by adding a bounded per-dimension scale that tames embedding-layer gradients, cutting memory while raising perplexity performance."}],"snapshot_sha256":"9959ef4638bbf38b963afe1a2259eda47df1e90fb591bc7e06d3b07a3a633033"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"d2e5cd0875624164da59059db0f1aca195c89aacecf5a84d78bc5af301a8de37"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.07663/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"The AdamW optimizer, while standard for LLM pretraining, is a critical memory bottleneck, consuming optimizer states equivalent to twice the model's size. Although light-state optimizers like SinkGD attempt to address this issue, we identify the embedding layer dilemma: these methods fail to handle the sparse, high-variance gradients inherent to embeddings, forcing a hybrid design that reverts to AdamW and partially negates the memory gains. We propose SAGE (Sign Adaptive GradiEnt), a novel optimizer that resolves this dilemma by replacing AdamW in this hybrid structure. SAGE combines a Lion-s","authors_text":"Hyun-Tae Kim, Wooin Lee","cross_cats":[],"headline":"SAGE replaces AdamW in hybrid LLM optimizers by adding a bounded per-dimension scale that tames embedding-layer gradients, cutting memory while raising perplexity performance.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-09T00:07:38Z","title":"SAGE: Sign-Adaptive Gradient for Memory-Efficient LLM Optimization"},"references":{"count":3,"internal_anchors":1,"resolved_work":3,"sample":[{"cited_arxiv_id":"1607.06450","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Layer Normalization","work_id":"20a2d720-0046-4c7c-bcd6-327ec8143f69","year":2016},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"DAtts L ⊙ 1√ h softmax′ gAtt s L√ h !# Y K L , DY K L =","work_id":"fe712d95-0ed1-4d1a-94aa-b36a44a56919","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Opt.\" refers to optimizer state memory, while","work_id":"68f80759-6f11-4808-8e11-3247f4a03dd9","year":2048}],"snapshot_sha256":"2577a907f5d50e501c1dd2a2c777f4dce9b45a3f6abf1b802efbb7184d046c09"},"source":{"id":"2604.07663","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-10T18:33:19.243230Z","id":"b0946ce6-2077-4e74-877c-38f3fb057507","model_set":{"reader":"grok-4.3"},"one_line_summary":"SAGE replaces AdamW in memory-efficient LLM hybrids with a Lion-style sign update plus a provably bounded O(d) adaptive scale, delivering SOTA perplexity on Llama-1.3B while cutting optimizer-state memory.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"SAGE replaces AdamW in hybrid LLM optimizers by adding a bounded per-dimension scale that tames embedding-layer gradients, cutting memory while raising perplexity performance.","strongest_claim":"On Llama models up to 1.3B parameters, our SAGE-based hybrid achieves new state-of-the-art perplexity, outperforming all baselines, including SinkGD hybrid, while significantly reducing optimizer state memory.","weakest_assumption":"That the high-variance, sparse gradients in the embedding layer can be sufficiently controlled by a single per-dimension scale provably bounded by 1.0 without reintroducing the instability that forced prior methods back to AdamW."}},"verdict_id":"b0946ce6-2077-4e74-877c-38f3fb057507"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:e30b2641b931d75633f42e6aa241a18f470145e81461f469e8ebd6ef46cd0023","target":"record","created_at":"2026-06-26T01:15:51Z","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":"5a8f77f162041ef977b19efbf3b0e41f621029f50f0f7f23836c3f9a532a3e23","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-09T00:07:38Z","title_canon_sha256":"aa16abe577c9cc3b4f2638308385ac68083532f098b0c3f4f4a2f80d4b295b19"},"schema_version":"1.0","source":{"id":"2604.07663","kind":"arxiv","version":2}},"canonical_sha256":"540d0d9e85a8fb04de63cd238d38c2f2182086c8d712ff6ee20cb40ffaa9e74d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"540d0d9e85a8fb04de63cd238d38c2f2182086c8d712ff6ee20cb40ffaa9e74d","first_computed_at":"2026-06-26T01:15:51.382011Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-26T01:15:51.382011Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"YEFqOUzepbftRQ3FGXbPtW4Ax9j3Nh+O8QJweaC/e5JHlrIBeDnpIdAygtVWg7LTAAFaBhh3Y+EKldUvNG+uCA==","signature_status":"signed_v1","signed_at":"2026-06-26T01:15:51.382405Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.07663","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:e30b2641b931d75633f42e6aa241a18f470145e81461f469e8ebd6ef46cd0023","sha256:2264cac6a464e4a589ed77b48b6c92394f92a8e8332ba9f7e52c50625a5f31f2"],"state_sha256":"17892e9bf4859fe5f1e2ee683725732b30c69341c144529272cb6fe8cce5bb96"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"S+pK/jMzQo+/EL43SKi+OCFZQMKnOJJJOJUxR6doJAZyraNTnyuiO2VZVgyMi0WvB0SD2ASnO0tY7NHVcB2pCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-26T04:38:55.089673Z","bundle_sha256":"311d4e4990e68af2e5d4c12cc34b10021f57662e3dc8d6f52f9750981134a3b6"}}