{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Q6O45TFZHJOJMTTE2U3XZPOT2Y","short_pith_number":"pith:Q6O45TFZ","schema_version":"1.0","canonical_sha256":"879dceccb93a5c964e64d5377cbdd3d604a63b9492eca36b415f29ee7b66c594","source":{"kind":"arxiv","id":"2505.14136","version":2},"attestation_state":"computed","paper":{"title":"Local Mixtures of Experts: Essentially Free Test-Time Training via Model Merging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Andreas Krause, Ido Hakimi, Jonas H\\\"ubotter, Ryo Bertolissi","submitted_at":"2025-05-20T09:39:54Z","abstract_excerpt":"Mixture of expert (MoE) models are a promising approach to increasing model capacity without increasing inference cost, and are core components of many state-of-the-art language models. However, current MoE models typically use only few experts due to prohibitive training and inference cost. We propose Test-Time Model Merging (TTMM) which scales the MoE paradigm to an order of magnitude more experts and uses model merging to avoid almost any test-time overhead. We show that TTMM is an approximation of test-time training (TTT), which fine-tunes an expert model for each prediction task, i.e., pr"},"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":"2505.14136","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-20T09:39:54Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"30c5a2dff7ffd56a729da506bc36fc842a43f846f574a77bdf1f3325c1dfdb48","abstract_canon_sha256":"17c303555ccfa3f9923d62a1f7eb6cc6070f86b06986eb4fe25bcb6400de6e56"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:45:35.798942Z","signature_b64":"LT7FCJQpB9u2EgkdhYdVThPXAWwDMdf0+QIxwTr6cG2RlDR70TiBQolaaJ9cpbF7dDbSjas5y5DNlgtvIak8CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"879dceccb93a5c964e64d5377cbdd3d604a63b9492eca36b415f29ee7b66c594","last_reissued_at":"2026-07-05T11:45:35.798423Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:45:35.798423Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Local Mixtures of Experts: Essentially Free Test-Time Training via Model Merging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Andreas Krause, Ido Hakimi, Jonas H\\\"ubotter, Ryo Bertolissi","submitted_at":"2025-05-20T09:39:54Z","abstract_excerpt":"Mixture of expert (MoE) models are a promising approach to increasing model capacity without increasing inference cost, and are core components of many state-of-the-art language models. However, current MoE models typically use only few experts due to prohibitive training and inference cost. We propose Test-Time Model Merging (TTMM) which scales the MoE paradigm to an order of magnitude more experts and uses model merging to avoid almost any test-time overhead. We show that TTMM is an approximation of test-time training (TTT), which fine-tunes an expert model for each prediction task, i.e., pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.14136","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/2505.14136/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":"2505.14136","created_at":"2026-07-05T11:45:35.798485+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.14136v2","created_at":"2026-07-05T11:45:35.798485+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.14136","created_at":"2026-07-05T11:45:35.798485+00:00"},{"alias_kind":"pith_short_12","alias_value":"Q6O45TFZHJOJ","created_at":"2026-07-05T11:45:35.798485+00:00"},{"alias_kind":"pith_short_16","alias_value":"Q6O45TFZHJOJMTTE","created_at":"2026-07-05T11:45:35.798485+00:00"},{"alias_kind":"pith_short_8","alias_value":"Q6O45TFZ","created_at":"2026-07-05T11:45:35.798485+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.20189","citing_title":"SOLAR: A Self-Optimizing Open-Ended Autonomous Agent for Lifelong Learning and Continual Adaptation","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2507.18809","citing_title":"Test-time Offline Reinforcement Learning on Goal-related Experience","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2603.02945","citing_title":"ACE-Merging: Data-Free Model Merging with Adaptive Covariance Estimation","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y","json":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y.json","graph_json":"https://pith.science/api/pith-number/Q6O45TFZHJOJMTTE2U3XZPOT2Y/graph.json","events_json":"https://pith.science/api/pith-number/Q6O45TFZHJOJMTTE2U3XZPOT2Y/events.json","paper":"https://pith.science/paper/Q6O45TFZ"},"agent_actions":{"view_html":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y","download_json":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y.json","view_paper":"https://pith.science/paper/Q6O45TFZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.14136&json=true","fetch_graph":"https://pith.science/api/pith-number/Q6O45TFZHJOJMTTE2U3XZPOT2Y/graph.json","fetch_events":"https://pith.science/api/pith-number/Q6O45TFZHJOJMTTE2U3XZPOT2Y/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y/action/storage_attestation","attest_author":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y/action/author_attestation","sign_citation":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y/action/citation_signature","submit_replication":"https://pith.science/pith/Q6O45TFZHJOJMTTE2U3XZPOT2Y/action/replication_record"}},"created_at":"2026-07-05T11:45:35.798485+00:00","updated_at":"2026-07-05T11:45:35.798485+00:00"}