{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:7RDKHMXPEWZIE43VSUQCFWU2U3","short_pith_number":"pith:7RDKHMXP","schema_version":"1.0","canonical_sha256":"fc46a3b2ef25b2827375952022da9aa6fb1ebc1b02f6e085a5fba2b420e2c34f","source":{"kind":"arxiv","id":"2312.07987","version":3},"attestation_state":"computed","paper":{"title":"SwitchHead: Accelerating Transformers with Mixture-of-Experts Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.NE"],"primary_cat":"cs.LG","authors_text":"J\\\"urgen Schmidhuber, Kazuki Irie, Piotr Pi\\k{e}kos, R\\'obert Csord\\'as","submitted_at":"2023-12-13T09:00:21Z","abstract_excerpt":"Despite many recent works on Mixture of Experts (MoEs) for resource-efficient Transformer language models, existing methods mostly focus on MoEs for feedforward layers. Previous attempts at extending MoE to the self-attention layer fail to match the performance of the parameter-matched baseline. Our novel SwitchHead is an effective MoE method for the attention layer that successfully reduces both the compute and memory requirements, achieving wall-clock speedup, while matching the language modeling performance of the baseline Transformer. Our novel MoE mechanism allows SwitchHead to compute up"},"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":"2312.07987","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-12-13T09:00:21Z","cross_cats_sorted":["cs.CL","cs.NE"],"title_canon_sha256":"68edd1439cfb4acac4535ea14eafa1060444bbf7209cc30b9056bb46790a3b8d","abstract_canon_sha256":"b28c3b986dfdf9c9428add5e209f1f67932ba4b475dd290105d55db696d6f3b9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:13:48.719992Z","signature_b64":"qMqRseLjyx9D7lzvETFDJsDpqV4CKFdcd16G9pe1huSFpmoge9u2WolACSfXwgKPbjwd90xS6l+j1hUKRIbdDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fc46a3b2ef25b2827375952022da9aa6fb1ebc1b02f6e085a5fba2b420e2c34f","last_reissued_at":"2026-07-05T09:13:48.719532Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:13:48.719532Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SwitchHead: Accelerating Transformers with Mixture-of-Experts Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.NE"],"primary_cat":"cs.LG","authors_text":"J\\\"urgen Schmidhuber, Kazuki Irie, Piotr Pi\\k{e}kos, R\\'obert Csord\\'as","submitted_at":"2023-12-13T09:00:21Z","abstract_excerpt":"Despite many recent works on Mixture of Experts (MoEs) for resource-efficient Transformer language models, existing methods mostly focus on MoEs for feedforward layers. Previous attempts at extending MoE to the self-attention layer fail to match the performance of the parameter-matched baseline. Our novel SwitchHead is an effective MoE method for the attention layer that successfully reduces both the compute and memory requirements, achieving wall-clock speedup, while matching the language modeling performance of the baseline Transformer. Our novel MoE mechanism allows SwitchHead to compute up"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.07987","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/2312.07987/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":"2312.07987","created_at":"2026-07-05T09:13:48.719586+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.07987v3","created_at":"2026-07-05T09:13:48.719586+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.07987","created_at":"2026-07-05T09:13:48.719586+00:00"},{"alias_kind":"pith_short_12","alias_value":"7RDKHMXPEWZI","created_at":"2026-07-05T09:13:48.719586+00:00"},{"alias_kind":"pith_short_16","alias_value":"7RDKHMXPEWZIE43V","created_at":"2026-07-05T09:13:48.719586+00:00"},{"alias_kind":"pith_short_8","alias_value":"7RDKHMXP","created_at":"2026-07-05T09:13:48.719586+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06601","citing_title":"TriRoute: Unified Learned Routing for Joint Adaptive Attention, Experts, and KV-Cache Allocation","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09165","citing_title":"Sparse Layers are Critical to Scaling Looped Language Models","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09165","citing_title":"Sparse Layers are Critical to Scaling Looped Language Models","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3","json":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3.json","graph_json":"https://pith.science/api/pith-number/7RDKHMXPEWZIE43VSUQCFWU2U3/graph.json","events_json":"https://pith.science/api/pith-number/7RDKHMXPEWZIE43VSUQCFWU2U3/events.json","paper":"https://pith.science/paper/7RDKHMXP"},"agent_actions":{"view_html":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3","download_json":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3.json","view_paper":"https://pith.science/paper/7RDKHMXP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.07987&json=true","fetch_graph":"https://pith.science/api/pith-number/7RDKHMXPEWZIE43VSUQCFWU2U3/graph.json","fetch_events":"https://pith.science/api/pith-number/7RDKHMXPEWZIE43VSUQCFWU2U3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3/action/storage_attestation","attest_author":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3/action/author_attestation","sign_citation":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3/action/citation_signature","submit_replication":"https://pith.science/pith/7RDKHMXPEWZIE43VSUQCFWU2U3/action/replication_record"}},"created_at":"2026-07-05T09:13:48.719586+00:00","updated_at":"2026-07-05T09:13:48.719586+00:00"}