{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DTIS3R4SNP2OKSIOHAAVEFI3GM","short_pith_number":"pith:DTIS3R4S","schema_version":"1.0","canonical_sha256":"1cd12dc7926bf4e5490e380152151b3319b8ee24b8a52b5083e69c6b7e046584","source":{"kind":"arxiv","id":"2402.03804","version":1},"attestation_state":"computed","paper":{"title":"ReLU$^2$ Wins: Discovering Efficient Activation Functions for Sparse LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chaojun Xiao, Chenyang Song, Guanghui Yu, Maosong Sun, Xu Han, Yankai Lin, Yixin Song, Zeyu Mi, Zhengyan Zhang, Zhiyuan Liu","submitted_at":"2024-02-06T08:45:51Z","abstract_excerpt":"Sparse computation offers a compelling solution for the inference of Large Language Models (LLMs) in low-resource scenarios by dynamically skipping the computation of inactive neurons. While traditional approaches focus on ReLU-based LLMs, leveraging zeros in activation values, we broaden the scope of sparse LLMs beyond zero activation values. We introduce a general method that defines neuron activation through neuron output magnitudes and a tailored magnitude threshold, demonstrating that non-ReLU LLMs also exhibit sparse activation. To find the most efficient activation function for sparse c"},"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":"2402.03804","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-02-06T08:45:51Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7cd6f52e2b3613c7a51e6d5c6d257b1ceb412abdbde9d8bc4141c31b98d0d64b","abstract_canon_sha256":"4e6974c50d641a8935c4d44308c3f96bcbeea0f7e196a4efd9f9987f3a9d55f0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:41:58.982458Z","signature_b64":"5pKCEjWf7uq/brjxYcRTJmdmvwqfxLH7HNNrboufHncC4TbIH+dxFU+HbeAQm4DvRzV523gpaCsIXxMrQKGUDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1cd12dc7926bf4e5490e380152151b3319b8ee24b8a52b5083e69c6b7e046584","last_reissued_at":"2026-07-05T07:41:58.982095Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:41:58.982095Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ReLU$^2$ Wins: Discovering Efficient Activation Functions for Sparse LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chaojun Xiao, Chenyang Song, Guanghui Yu, Maosong Sun, Xu Han, Yankai Lin, Yixin Song, Zeyu Mi, Zhengyan Zhang, Zhiyuan Liu","submitted_at":"2024-02-06T08:45:51Z","abstract_excerpt":"Sparse computation offers a compelling solution for the inference of Large Language Models (LLMs) in low-resource scenarios by dynamically skipping the computation of inactive neurons. While traditional approaches focus on ReLU-based LLMs, leveraging zeros in activation values, we broaden the scope of sparse LLMs beyond zero activation values. We introduce a general method that defines neuron activation through neuron output magnitudes and a tailored magnitude threshold, demonstrating that non-ReLU LLMs also exhibit sparse activation. To find the most efficient activation function for sparse c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.03804","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/2402.03804/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":"2402.03804","created_at":"2026-07-05T07:41:58.982153+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.03804v1","created_at":"2026-07-05T07:41:58.982153+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.03804","created_at":"2026-07-05T07:41:58.982153+00:00"},{"alias_kind":"pith_short_12","alias_value":"DTIS3R4SNP2O","created_at":"2026-07-05T07:41:58.982153+00:00"},{"alias_kind":"pith_short_16","alias_value":"DTIS3R4SNP2OKSIO","created_at":"2026-07-05T07:41:58.982153+00:00"},{"alias_kind":"pith_short_8","alias_value":"DTIS3R4S","created_at":"2026-07-05T07:41:58.982153+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00158","citing_title":"Readable but Not Controllable: Neuron-Level Evidence for Medical LLM Hallucination","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26647","citing_title":"More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26632","citing_title":"RT-Lynx: Putting the GEMM Sparsity In a Right Way for Diffusion Models","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23061","citing_title":"Anytime Training with Schedule-Free Spectral Optimization","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12744","citing_title":"Resting Neurons, Active Insights: Robustifying Activation Sparsity in LLMs via Spontaneity","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17659","citing_title":"Bug or Feature$^2$: Weight Drift, Activation Sparsity and Spikes","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17659","citing_title":"Bug or Feature$^2$: Weight Drift, Activation Sparsity and Spikes","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12744","citing_title":"Resting Neurons, Active Insights: Robustifying Activation Sparsity in LLMs via Spontaneity","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25150","citing_title":"The Role of Symmetry in Optimizing Overparameterized Networks","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24169","citing_title":"PointTransformerX: Portable and Efficient 3D Point Cloud Processing without Sparse Algorithms","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12946","citing_title":"Parcae: Scaling Laws For Stable Looped Language Models","ref_index":93,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25150","citing_title":"The Role of Symmetry in Optimizing Overparameterized Networks","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM","json":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM.json","graph_json":"https://pith.science/api/pith-number/DTIS3R4SNP2OKSIOHAAVEFI3GM/graph.json","events_json":"https://pith.science/api/pith-number/DTIS3R4SNP2OKSIOHAAVEFI3GM/events.json","paper":"https://pith.science/paper/DTIS3R4S"},"agent_actions":{"view_html":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM","download_json":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM.json","view_paper":"https://pith.science/paper/DTIS3R4S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.03804&json=true","fetch_graph":"https://pith.science/api/pith-number/DTIS3R4SNP2OKSIOHAAVEFI3GM/graph.json","fetch_events":"https://pith.science/api/pith-number/DTIS3R4SNP2OKSIOHAAVEFI3GM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM/action/storage_attestation","attest_author":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM/action/author_attestation","sign_citation":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM/action/citation_signature","submit_replication":"https://pith.science/pith/DTIS3R4SNP2OKSIOHAAVEFI3GM/action/replication_record"}},"created_at":"2026-07-05T07:41:58.982153+00:00","updated_at":"2026-07-05T07:41:58.982153+00:00"}