{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DEY3ZIF7L3WHUUOAT3CGLTKOXK","short_pith_number":"pith:DEY3ZIF7","schema_version":"1.0","canonical_sha256":"1931bca0bf5eec7a51c09ec465cd4ebab8405351b76cf8292d4cad8e1d3754cb","source":{"kind":"arxiv","id":"2405.03637","version":1},"attestation_state":"computed","paper":{"title":"Collage: Light-Weight Low-Precision Strategy for LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amith Mamidala, Anoop Deoras, Gaurav Gupta, Hao Zhou, Jeffrey Huynh, Karthick Gopalswamy, Luke Huan, Ron Diamant, Tao Yu, Youngsuk Park","submitted_at":"2024-05-06T16:55:30Z","abstract_excerpt":"Large models training is plagued by the intense compute cost and limited hardware memory. A practical solution is low-precision representation but is troubled by loss in numerical accuracy and unstable training rendering the model less useful. We argue that low-precision floating points can perform well provided the error is properly compensated at the critical locations in the training process. We propose Collage which utilizes multi-component float representation in low-precision to accurately perform operations with numerical errors accounted. To understand the impact of imprecision to trai"},"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":"2405.03637","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-06T16:55:30Z","cross_cats_sorted":[],"title_canon_sha256":"dd052fabd72c3f280021b543f28f16ab06e265fb21566873c7bbcdba9d4d127d","abstract_canon_sha256":"151a40ecfa440783031607466d1fc9bd61d996682d15437f7a6ac43e07cfb620"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:16:05.623588Z","signature_b64":"GE4c5zLUteeEZyPkTWF/JNGoO1rMS1nrdMFCpnre4bYTkmfvFB5HU6qChlGeKkP3VAOUjGp/nOacG/xlKV2ZCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1931bca0bf5eec7a51c09ec465cd4ebab8405351b76cf8292d4cad8e1d3754cb","last_reissued_at":"2026-07-05T08:16:05.623199Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:16:05.623199Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Collage: Light-Weight Low-Precision Strategy for LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amith Mamidala, Anoop Deoras, Gaurav Gupta, Hao Zhou, Jeffrey Huynh, Karthick Gopalswamy, Luke Huan, Ron Diamant, Tao Yu, Youngsuk Park","submitted_at":"2024-05-06T16:55:30Z","abstract_excerpt":"Large models training is plagued by the intense compute cost and limited hardware memory. A practical solution is low-precision representation but is troubled by loss in numerical accuracy and unstable training rendering the model less useful. We argue that low-precision floating points can perform well provided the error is properly compensated at the critical locations in the training process. We propose Collage which utilizes multi-component float representation in low-precision to accurately perform operations with numerical errors accounted. To understand the impact of imprecision to trai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.03637","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/2405.03637/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":"2405.03637","created_at":"2026-07-05T08:16:05.623249+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.03637v1","created_at":"2026-07-05T08:16:05.623249+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.03637","created_at":"2026-07-05T08:16:05.623249+00:00"},{"alias_kind":"pith_short_12","alias_value":"DEY3ZIF7L3WH","created_at":"2026-07-05T08:16:05.623249+00:00"},{"alias_kind":"pith_short_16","alias_value":"DEY3ZIF7L3WHUUOA","created_at":"2026-07-05T08:16:05.623249+00:00"},{"alias_kind":"pith_short_8","alias_value":"DEY3ZIF7","created_at":"2026-07-05T08:16:05.623249+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/DEY3ZIF7L3WHUUOAT3CGLTKOXK","json":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK.json","graph_json":"https://pith.science/api/pith-number/DEY3ZIF7L3WHUUOAT3CGLTKOXK/graph.json","events_json":"https://pith.science/api/pith-number/DEY3ZIF7L3WHUUOAT3CGLTKOXK/events.json","paper":"https://pith.science/paper/DEY3ZIF7"},"agent_actions":{"view_html":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK","download_json":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK.json","view_paper":"https://pith.science/paper/DEY3ZIF7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.03637&json=true","fetch_graph":"https://pith.science/api/pith-number/DEY3ZIF7L3WHUUOAT3CGLTKOXK/graph.json","fetch_events":"https://pith.science/api/pith-number/DEY3ZIF7L3WHUUOAT3CGLTKOXK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK/action/storage_attestation","attest_author":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK/action/author_attestation","sign_citation":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK/action/citation_signature","submit_replication":"https://pith.science/pith/DEY3ZIF7L3WHUUOAT3CGLTKOXK/action/replication_record"}},"created_at":"2026-07-05T08:16:05.623249+00:00","updated_at":"2026-07-05T08:16:05.623249+00:00"}