{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PSHTPY52PXVIIYEGNDAN3M7QPK","short_pith_number":"pith:PSHTPY52","schema_version":"1.0","canonical_sha256":"7c8f37e3ba7dea84608668c0ddb3f07a8120cf832c35d8a77c07e4c53f2ca9b0","source":{"kind":"arxiv","id":"2402.13388","version":3},"attestation_state":"computed","paper":{"title":"Transformer tricks: Precomputing the first layer","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Nils Graef","submitted_at":"2024-02-20T21:34:56Z","abstract_excerpt":"This micro-paper describes a trick to speed up inference of transformers with RoPE (such as LLaMA, Mistral, PaLM, and Gemma). For these models, a large portion of the first transformer layer can be precomputed, which results in slightly lower latency and lower cost-per-token. Because this trick optimizes only one layer, the relative savings depend on the total number of layers. For example, the maximum savings for a model with only 4 layers (such as Whisper tiny) is limited to 25%, while a 32-layer model is limited to 3% savings. See https://github.com/OpenMachine-ai/transformer-tricks for cod"},"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":"2402.13388","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-20T21:34:56Z","cross_cats_sorted":[],"title_canon_sha256":"eb4a9b6850d289872aaf88e9a2ca3de3604372d65bac42c806a9b0537b9511e8","abstract_canon_sha256":"f3ca8c4284260c651f5c522fc46de4233f042161470827e155fbc05d803ed741"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:54:55.000257Z","signature_b64":"GOavLJ//hsmqnHoXhgvpHukh2g85Q172DyPfwSE2Ts44WJI5hgUsTMKEq9N5VUqiiQ9R6deVZbJYvpJz7jeZDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7c8f37e3ba7dea84608668c0ddb3f07a8120cf832c35d8a77c07e4c53f2ca9b0","last_reissued_at":"2026-07-05T07:54:54.999850Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:54:54.999850Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformer tricks: Precomputing the first layer","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Nils Graef","submitted_at":"2024-02-20T21:34:56Z","abstract_excerpt":"This micro-paper describes a trick to speed up inference of transformers with RoPE (such as LLaMA, Mistral, PaLM, and Gemma). For these models, a large portion of the first transformer layer can be precomputed, which results in slightly lower latency and lower cost-per-token. Because this trick optimizes only one layer, the relative savings depend on the total number of layers. For example, the maximum savings for a model with only 4 layers (such as Whisper tiny) is limited to 25%, while a 32-layer model is limited to 3% savings. See https://github.com/OpenMachine-ai/transformer-tricks for cod"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.13388","kind":"arxiv","version":3},"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/2402.13388/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":"2402.13388","created_at":"2026-07-05T07:54:54.999907+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.13388v3","created_at":"2026-07-05T07:54:54.999907+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.13388","created_at":"2026-07-05T07:54:54.999907+00:00"},{"alias_kind":"pith_short_12","alias_value":"PSHTPY52PXVI","created_at":"2026-07-05T07:54:54.999907+00:00"},{"alias_kind":"pith_short_16","alias_value":"PSHTPY52PXVIIYEG","created_at":"2026-07-05T07:54:54.999907+00:00"},{"alias_kind":"pith_short_8","alias_value":"PSHTPY52","created_at":"2026-07-05T07:54:54.999907+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2407.09577","citing_title":"FlashNorm: Fast Normalization for Transformers","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK","json":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK.json","graph_json":"https://pith.science/api/pith-number/PSHTPY52PXVIIYEGNDAN3M7QPK/graph.json","events_json":"https://pith.science/api/pith-number/PSHTPY52PXVIIYEGNDAN3M7QPK/events.json","paper":"https://pith.science/paper/PSHTPY52"},"agent_actions":{"view_html":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK","download_json":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK.json","view_paper":"https://pith.science/paper/PSHTPY52","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.13388&json=true","fetch_graph":"https://pith.science/api/pith-number/PSHTPY52PXVIIYEGNDAN3M7QPK/graph.json","fetch_events":"https://pith.science/api/pith-number/PSHTPY52PXVIIYEGNDAN3M7QPK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK/action/storage_attestation","attest_author":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK/action/author_attestation","sign_citation":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK/action/citation_signature","submit_replication":"https://pith.science/pith/PSHTPY52PXVIIYEGNDAN3M7QPK/action/replication_record"}},"created_at":"2026-07-05T07:54:54.999907+00:00","updated_at":"2026-07-05T07:54:54.999907+00:00"}