{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:7TN4LDUIK53ANGHLCDV353HZCU","short_pith_number":"pith:7TN4LDUI","schema_version":"1.0","canonical_sha256":"fcdbc58e8857760698eb10ebbeecf9153af746bcd916fdc7e1b4929ce4129182","source":{"kind":"arxiv","id":"2402.07871","version":1},"attestation_state":"computed","paper":{"title":"Scaling Laws for Fine-Grained Mixture of Experts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jakub Krajewski, Jan Ludziejewski, Kamil Adamczewski, Kamil Ciebiera, Krystian Kr\\'ol, Maciej Pi\\'oro, Marek Cygan, Micha{\\l} Krutul, Piotr Sankowski, Sebastian Jaszczur, Szymon Antoniak, Tomasz Odrzyg\\'o\\'zd\\'z","submitted_at":"2024-02-12T18:33:47Z","abstract_excerpt":"Mixture of Experts (MoE) models have emerged as a primary solution for reducing the computational cost of Large Language Models. In this work, we analyze their scaling properties, incorporating an expanded range of variables. Specifically, we introduce a new hyperparameter, granularity, whose adjustment enables precise control over the size of the experts. Building on this, we establish scaling laws for fine-grained MoE, taking into account the number of training tokens, model size, and granularity. Leveraging these laws, we derive the optimal training configuration for a given computational b"},"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":"2402.07871","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-02-12T18:33:47Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"93c97d780071f155cc084a5326ecff132cab7746b1aea2666a386872819ebbe8","abstract_canon_sha256":"e5c9cc1192d45b36808cc718fd14d24e24ded363db4246d3ec024a2beadc1432"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:44:13.520447Z","signature_b64":"rVBDJry3g4FCQHHoT+3/Nd9FiAFpgpTkFCz7BUvBkgxxIrRRUIFRji+o+YvaV9YoeGlUl1VJyoM9AFWscUVbBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fcdbc58e8857760698eb10ebbeecf9153af746bcd916fdc7e1b4929ce4129182","last_reissued_at":"2026-07-05T07:44:13.519928Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:44:13.519928Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scaling Laws for Fine-Grained Mixture of Experts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jakub Krajewski, Jan Ludziejewski, Kamil Adamczewski, Kamil Ciebiera, Krystian Kr\\'ol, Maciej Pi\\'oro, Marek Cygan, Micha{\\l} Krutul, Piotr Sankowski, Sebastian Jaszczur, Szymon Antoniak, Tomasz Odrzyg\\'o\\'zd\\'z","submitted_at":"2024-02-12T18:33:47Z","abstract_excerpt":"Mixture of Experts (MoE) models have emerged as a primary solution for reducing the computational cost of Large Language Models. In this work, we analyze their scaling properties, incorporating an expanded range of variables. Specifically, we introduce a new hyperparameter, granularity, whose adjustment enables precise control over the size of the experts. Building on this, we establish scaling laws for fine-grained MoE, taking into account the number of training tokens, model size, and granularity. Leveraging these laws, we derive the optimal training configuration for a given computational b"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.07871","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/2402.07871/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":"2402.07871","created_at":"2026-07-05T07:44:13.519992+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.07871v1","created_at":"2026-07-05T07:44:13.519992+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.07871","created_at":"2026-07-05T07:44:13.519992+00:00"},{"alias_kind":"pith_short_12","alias_value":"7TN4LDUIK53A","created_at":"2026-07-05T07:44:13.519992+00:00"},{"alias_kind":"pith_short_16","alias_value":"7TN4LDUIK53ANGHL","created_at":"2026-07-05T07:44:13.519992+00:00"},{"alias_kind":"pith_short_8","alias_value":"7TN4LDUI","created_at":"2026-07-05T07:44:13.519992+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":25,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21228","citing_title":"Sakana Fugu Technical Report","ref_index":197,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07414","citing_title":"Sparsely gated tiny linear experts","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24037","citing_title":"A Limit Theory of Foundation Models: A Mathematical Approach to Understanding Emergent Intelligence and Scaling Laws","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31397","citing_title":"Mixture-of-Control: State-Aware Fine-Tuning for Transformer-based Models","ref_index":98,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27358","citing_title":"MobileMoE: Scaling On-Device Mixture of Experts","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27722","citing_title":"NUCLEUS-MoE: Unified Model of Pool Boiling for Liquid Cooling","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31268","citing_title":"Mellum2 Technical Report","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":129,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16349","citing_title":"Geometric Asymmetry in MoE Specialization: Functional Decorrelation and Representational Overlap","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12715","citing_title":"Scaling Laws for Mixture Pretraining Under Data Constraints","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2507.00029","citing_title":"LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18245","citing_title":"Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2509.19349","citing_title":"ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution","ref_index":163,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14200","citing_title":"How to Scale Mixture-of-Experts: From muP to the Maximally Scale-Stable Parameterization","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12715","citing_title":"Scaling Laws for Mixture Pretraining Under Data Constraints","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13769","citing_title":"Dense vs Sparse Pretraining at Tiny Scale: Active-Parameter vs Total-Parameter Matching","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":129,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24037","citing_title":"A Limit Theory of Foundation Models: A Mathematical Approach to Understanding Emergent Intelligence and Scaling Laws","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":129,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19835","citing_title":"Expert Upcycling: Shifting the Compute-Efficient Frontier of Mixture-of-Experts","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06665","citing_title":"UniPool: A Globally Shared Expert Pool for Mixture-of-Experts","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21691","citing_title":"There Will Be a Scientific Theory of Deep Learning","ref_index":125,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09175","citing_title":"Generalization and Scaling Laws for Mixture-of-Experts Transformers","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18473","citing_title":"Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19835","citing_title":"Expert Upcycling: Shifting the Compute-Efficient Frontier of Mixture-of-Experts","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU","json":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU.json","graph_json":"https://pith.science/api/pith-number/7TN4LDUIK53ANGHLCDV353HZCU/graph.json","events_json":"https://pith.science/api/pith-number/7TN4LDUIK53ANGHLCDV353HZCU/events.json","paper":"https://pith.science/paper/7TN4LDUI"},"agent_actions":{"view_html":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU","download_json":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU.json","view_paper":"https://pith.science/paper/7TN4LDUI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.07871&json=true","fetch_graph":"https://pith.science/api/pith-number/7TN4LDUIK53ANGHLCDV353HZCU/graph.json","fetch_events":"https://pith.science/api/pith-number/7TN4LDUIK53ANGHLCDV353HZCU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU/action/storage_attestation","attest_author":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU/action/author_attestation","sign_citation":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU/action/citation_signature","submit_replication":"https://pith.science/pith/7TN4LDUIK53ANGHLCDV353HZCU/action/replication_record"}},"created_at":"2026-07-05T07:44:13.519992+00:00","updated_at":"2026-07-05T07:44:13.519992+00:00"}