{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:IN4IDGDTCUQPSNSJY7GEEL76CB","short_pith_number":"pith:IN4IDGDT","schema_version":"1.0","canonical_sha256":"43788198731520f93649c7cc422ffe10544253ceb68d47dfe5e01d10f9583f8b","source":{"kind":"arxiv","id":"2402.10930","version":3},"attestation_state":"computed","paper":{"title":"ConSmax: Hardware-Friendly Alternative Softmax with Learnable Parameters","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.AR","authors_text":"Bangfei Pan, Dennis Sylvester, Derek Chow, Gregory Kielian, Guanchen Tao, Kauna Lei, Mehdi Saligane, Shiwei Liu, Yifei Zou, Zichen Fan","submitted_at":"2024-01-31T17:52:52Z","abstract_excerpt":"The self-attention mechanism distinguishes transformer-based large language models (LLMs) apart from convolutional and recurrent neural networks. Despite the performance improvement, achieving real-time LLM inference on silicon remains challenging due to the extensive use of Softmax in self-attention. In addition to the non-linearity, the low arithmetic intensity significantly limits processing parallelism, especially when working with longer contexts. To address this challenge, we propose Constant Softmax (ConSmax), a software-hardware co-design that serves as an efficient alternative to Soft"},"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":"2402.10930","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AR","submitted_at":"2024-01-31T17:52:52Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"701e801dce8ba250d84dd6d41bca2e689c00630c37a05e76f867e5cbe3693086","abstract_canon_sha256":"ebbe38399457d82809d04591ce5eb33d5a572ca80ba7ba4f0e00c0467618025a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:35:32.390784Z","signature_b64":"2wcJwk8NGUITvHTbEhfLokM8QHkAw02fH3KJwLwNO+CUl7LXvAW5zFTOu2g40jFb/IrCqAMq4AHqklTqKs+jDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43788198731520f93649c7cc422ffe10544253ceb68d47dfe5e01d10f9583f8b","last_reissued_at":"2026-07-05T09:35:32.390329Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:35:32.390329Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ConSmax: Hardware-Friendly Alternative Softmax with Learnable Parameters","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.AR","authors_text":"Bangfei Pan, Dennis Sylvester, Derek Chow, Gregory Kielian, Guanchen Tao, Kauna Lei, Mehdi Saligane, Shiwei Liu, Yifei Zou, Zichen Fan","submitted_at":"2024-01-31T17:52:52Z","abstract_excerpt":"The self-attention mechanism distinguishes transformer-based large language models (LLMs) apart from convolutional and recurrent neural networks. Despite the performance improvement, achieving real-time LLM inference on silicon remains challenging due to the extensive use of Softmax in self-attention. In addition to the non-linearity, the low arithmetic intensity significantly limits processing parallelism, especially when working with longer contexts. To address this challenge, we propose Constant Softmax (ConSmax), a software-hardware co-design that serves as an efficient alternative to Soft"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.10930","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/2402.10930/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":"2402.10930","created_at":"2026-07-05T09:35:32.390380+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.10930v3","created_at":"2026-07-05T09:35:32.390380+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.10930","created_at":"2026-07-05T09:35:32.390380+00:00"},{"alias_kind":"pith_short_12","alias_value":"IN4IDGDTCUQP","created_at":"2026-07-05T09:35:32.390380+00:00"},{"alias_kind":"pith_short_16","alias_value":"IN4IDGDTCUQPSNSJ","created_at":"2026-07-05T09:35:32.390380+00:00"},{"alias_kind":"pith_short_8","alias_value":"IN4IDGDT","created_at":"2026-07-05T09:35:32.390380+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.14201","citing_title":"FLASH-D: FlashAttention with Hidden Softmax Division","ref_index":22,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB","json":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB.json","graph_json":"https://pith.science/api/pith-number/IN4IDGDTCUQPSNSJY7GEEL76CB/graph.json","events_json":"https://pith.science/api/pith-number/IN4IDGDTCUQPSNSJY7GEEL76CB/events.json","paper":"https://pith.science/paper/IN4IDGDT"},"agent_actions":{"view_html":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB","download_json":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB.json","view_paper":"https://pith.science/paper/IN4IDGDT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.10930&json=true","fetch_graph":"https://pith.science/api/pith-number/IN4IDGDTCUQPSNSJY7GEEL76CB/graph.json","fetch_events":"https://pith.science/api/pith-number/IN4IDGDTCUQPSNSJY7GEEL76CB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB/action/storage_attestation","attest_author":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB/action/author_attestation","sign_citation":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB/action/citation_signature","submit_replication":"https://pith.science/pith/IN4IDGDTCUQPSNSJY7GEEL76CB/action/replication_record"}},"created_at":"2026-07-05T09:35:32.390380+00:00","updated_at":"2026-07-05T09:35:32.390380+00:00"}