{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VZ7CHIA3YS6JJKZ3F4W3AO44WM","short_pith_number":"pith:VZ7CHIA3","schema_version":"1.0","canonical_sha256":"ae7e23a01bc4bc94ab3b2f2db03b9cb314c238cdf9d5ba3e1c50fa497bdd28aa","source":{"kind":"arxiv","id":"2410.15352","version":2},"attestation_state":"computed","paper":{"title":"CompAct: Compressed Activations for Memory-Efficient LLM Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Assaf Schuster, Nitzan Hodos, Yara Shamshoum, Yuval Sieradzki","submitted_at":"2024-10-20T10:24:38Z","abstract_excerpt":"We introduce CompAct, a technique that reduces peak memory utilization on GPU by 25-30% for pretraining and 50% for fine-tuning of LLMs. Peak device memory is a major limiting factor in training LLMs, with various recent works aiming to reduce model memory. However most works don't target the largest component of allocated memory during training: the model's compute graph, which is stored for the backward pass. By storing low-rank, compressed activations to be used in the backward pass we greatly reduce the required memory, unlike previous methods which only reduce optimizer overheads or the n"},"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":"2410.15352","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-20T10:24:38Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"10c660e19c582bff4b018c98b3d1af195b9c8b4af41f109ec60fffdd1e489793","abstract_canon_sha256":"d1896291a929ccf7187cf9af7f06420e2ed33280d00eb514a59a03f351a2c5d8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:11:26.765126Z","signature_b64":"+TCC4frEyJG2fSff4CBHlGNHjqzQaxp0ecHTI7gipGvf8tfSPkEy0OTBk8tgicFGgiGz1v6j7mb7sUryKDgKAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ae7e23a01bc4bc94ab3b2f2db03b9cb314c238cdf9d5ba3e1c50fa497bdd28aa","last_reissued_at":"2026-07-05T10:11:26.764618Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:11:26.764618Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CompAct: Compressed Activations for Memory-Efficient LLM Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Assaf Schuster, Nitzan Hodos, Yara Shamshoum, Yuval Sieradzki","submitted_at":"2024-10-20T10:24:38Z","abstract_excerpt":"We introduce CompAct, a technique that reduces peak memory utilization on GPU by 25-30% for pretraining and 50% for fine-tuning of LLMs. Peak device memory is a major limiting factor in training LLMs, with various recent works aiming to reduce model memory. However most works don't target the largest component of allocated memory during training: the model's compute graph, which is stored for the backward pass. By storing low-rank, compressed activations to be used in the backward pass we greatly reduce the required memory, unlike previous methods which only reduce optimizer overheads or the n"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.15352","kind":"arxiv","version":2},"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/2410.15352/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":"2410.15352","created_at":"2026-07-05T10:11:26.764681+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.15352v2","created_at":"2026-07-05T10:11:26.764681+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.15352","created_at":"2026-07-05T10:11:26.764681+00:00"},{"alias_kind":"pith_short_12","alias_value":"VZ7CHIA3YS6J","created_at":"2026-07-05T10:11:26.764681+00:00"},{"alias_kind":"pith_short_16","alias_value":"VZ7CHIA3YS6JJKZ3","created_at":"2026-07-05T10:11:26.764681+00:00"},{"alias_kind":"pith_short_8","alias_value":"VZ7CHIA3","created_at":"2026-07-05T10:11:26.764681+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2512.12131","citing_title":"BOOST: BOttleneck-Optimized Scalable Training Framework for Low-Rank Large Language Models","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM","json":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM.json","graph_json":"https://pith.science/api/pith-number/VZ7CHIA3YS6JJKZ3F4W3AO44WM/graph.json","events_json":"https://pith.science/api/pith-number/VZ7CHIA3YS6JJKZ3F4W3AO44WM/events.json","paper":"https://pith.science/paper/VZ7CHIA3"},"agent_actions":{"view_html":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM","download_json":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM.json","view_paper":"https://pith.science/paper/VZ7CHIA3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.15352&json=true","fetch_graph":"https://pith.science/api/pith-number/VZ7CHIA3YS6JJKZ3F4W3AO44WM/graph.json","fetch_events":"https://pith.science/api/pith-number/VZ7CHIA3YS6JJKZ3F4W3AO44WM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM/action/storage_attestation","attest_author":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM/action/author_attestation","sign_citation":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM/action/citation_signature","submit_replication":"https://pith.science/pith/VZ7CHIA3YS6JJKZ3F4W3AO44WM/action/replication_record"}},"created_at":"2026-07-05T10:11:26.764681+00:00","updated_at":"2026-07-05T10:11:26.764681+00:00"}