{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DATGRIXFOIJHMSLLBAQJ33IOOA","short_pith_number":"pith:DATGRIXF","schema_version":"1.0","canonical_sha256":"182668a2e5721276496b08209ded0e701f837f2b440b165178a876aba177f6d7","source":{"kind":"arxiv","id":"2405.03103","version":2},"attestation_state":"computed","paper":{"title":"Learning from Students: Applying t-Distributions to Explore Accurate and Efficient Formats for LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Bahaa Kotb, Gang Wu, Jordan Dotzel, Mohamed S. Abdelfattah, Sheng Li, Sushma Prasad, Yuzong Chen, Zhiru Zhang","submitted_at":"2024-05-06T01:39:59Z","abstract_excerpt":"The increasing size of large language models (LLMs) traditionally requires low-precision integer formats to meet strict latency and power demands. Yet recently, alternative formats such as Normal Float (NF4) have increased model accuracy at the cost of increased chip area. In this work, we first conduct a large-scale analysis of LLM weights and activations across 30 networks and conclude that most distributions follow a Student's t-distribution. We then derive a new theoretically optimal format, Student Float (SF4), that improves over NF4 across modern LLMs, for example increasing the average "},"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":"2405.03103","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-06T01:39:59Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"696af4c55f2cafe37375eaefa143347cbe1cbe051daac50425b34265a91016b6","abstract_canon_sha256":"01a1e2308081c1908e7d55e4e611b71c1ec3e9d32091e6246b8d787d3df24ecc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:29:46.388093Z","signature_b64":"616TfXtapfxzP3qevRzksRJ4BJYeDbtMXM0RfdDP46nzG9DFBmtR5VCUW4UOKPitzX3PvoqJ12qhgOHEzKa8DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"182668a2e5721276496b08209ded0e701f837f2b440b165178a876aba177f6d7","last_reissued_at":"2026-07-05T08:29:46.387626Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:29:46.387626Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning from Students: Applying t-Distributions to Explore Accurate and Efficient Formats for LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Bahaa Kotb, Gang Wu, Jordan Dotzel, Mohamed S. Abdelfattah, Sheng Li, Sushma Prasad, Yuzong Chen, Zhiru Zhang","submitted_at":"2024-05-06T01:39:59Z","abstract_excerpt":"The increasing size of large language models (LLMs) traditionally requires low-precision integer formats to meet strict latency and power demands. Yet recently, alternative formats such as Normal Float (NF4) have increased model accuracy at the cost of increased chip area. In this work, we first conduct a large-scale analysis of LLM weights and activations across 30 networks and conclude that most distributions follow a Student's t-distribution. We then derive a new theoretically optimal format, Student Float (SF4), that improves over NF4 across modern LLMs, for example increasing the average "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.03103","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/2405.03103/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":"2405.03103","created_at":"2026-07-05T08:29:46.387682+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.03103v2","created_at":"2026-07-05T08:29:46.387682+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.03103","created_at":"2026-07-05T08:29:46.387682+00:00"},{"alias_kind":"pith_short_12","alias_value":"DATGRIXFOIJH","created_at":"2026-07-05T08:29:46.387682+00:00"},{"alias_kind":"pith_short_16","alias_value":"DATGRIXFOIJHMSLL","created_at":"2026-07-05T08:29:46.387682+00:00"},{"alias_kind":"pith_short_8","alias_value":"DATGRIXF","created_at":"2026-07-05T08:29:46.387682+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/DATGRIXFOIJHMSLLBAQJ33IOOA","json":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA.json","graph_json":"https://pith.science/api/pith-number/DATGRIXFOIJHMSLLBAQJ33IOOA/graph.json","events_json":"https://pith.science/api/pith-number/DATGRIXFOIJHMSLLBAQJ33IOOA/events.json","paper":"https://pith.science/paper/DATGRIXF"},"agent_actions":{"view_html":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA","download_json":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA.json","view_paper":"https://pith.science/paper/DATGRIXF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.03103&json=true","fetch_graph":"https://pith.science/api/pith-number/DATGRIXFOIJHMSLLBAQJ33IOOA/graph.json","fetch_events":"https://pith.science/api/pith-number/DATGRIXFOIJHMSLLBAQJ33IOOA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA/action/storage_attestation","attest_author":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA/action/author_attestation","sign_citation":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA/action/citation_signature","submit_replication":"https://pith.science/pith/DATGRIXFOIJHMSLLBAQJ33IOOA/action/replication_record"}},"created_at":"2026-07-05T08:29:46.387682+00:00","updated_at":"2026-07-05T08:29:46.387682+00:00"}