{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XCFF2J62UNVQZ4GDOD2EVQG2XD","short_pith_number":"pith:XCFF2J62","schema_version":"1.0","canonical_sha256":"b88a5d27daa36b0cf0c370f44ac0dab8cfc32bab66e6f29ea8e467b2b3cecb8c","source":{"kind":"arxiv","id":"2501.11873","version":2},"attestation_state":"computed","paper":{"title":"Demons in the Detail: On Implementing Load Balancing Loss for Training Specialized Mixture-of-Expert Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Bo Zheng, Dayiheng Liu, Ivan Titov, Jingren Zhou, Junyang Lin, Kaiyue Wen, Rui Men, Zekun Wang, Zeyu Huang, Zihan Qiu","submitted_at":"2025-01-21T04:04:39Z","abstract_excerpt":"This paper revisits the implementation of $\\textbf{L}$oad-$\\textbf{b}$alancing $\\textbf{L}$oss (LBL) when training Mixture-of-Experts (MoEs) models. Specifically, LBL for MoEs is defined as $N_E \\sum_{i=1}^{N_E} f_i p_i$, where $N_E$ is the total number of experts, $f_i$ represents the frequency of expert $i$ being selected, and $p_i$ denotes the average gating score of the expert $i$. Existing MoE training frameworks usually employ the parallel training strategy so that $f_i$ and the LBL are calculated within a $\\textbf{micro-batch}$ and then averaged across parallel groups. In essence, a mic"},"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":"2501.11873","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-01-21T04:04:39Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"bd33c757eee372b6193233360361b00e6b869a6bdc8d2ebfe6cc4ce9e9958897","abstract_canon_sha256":"f11e59998d49a4dae978be40bebb90bd5c3eae9dc51e054b133a5abf48acb309"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:09:15.109337Z","signature_b64":"Vq8Ldh/NlMa/H1IP5MRCLlojI7DC9OvYQFfMqe5mcehaXmbVtX95bC3Cz3FdmV5aT0cSja47qcuosMUOaFt3Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b88a5d27daa36b0cf0c370f44ac0dab8cfc32bab66e6f29ea8e467b2b3cecb8c","last_reissued_at":"2026-07-05T10:09:15.108819Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:09:15.108819Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Demons in the Detail: On Implementing Load Balancing Loss for Training Specialized Mixture-of-Expert Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Bo Zheng, Dayiheng Liu, Ivan Titov, Jingren Zhou, Junyang Lin, Kaiyue Wen, Rui Men, Zekun Wang, Zeyu Huang, Zihan Qiu","submitted_at":"2025-01-21T04:04:39Z","abstract_excerpt":"This paper revisits the implementation of $\\textbf{L}$oad-$\\textbf{b}$alancing $\\textbf{L}$oss (LBL) when training Mixture-of-Experts (MoEs) models. Specifically, LBL for MoEs is defined as $N_E \\sum_{i=1}^{N_E} f_i p_i$, where $N_E$ is the total number of experts, $f_i$ represents the frequency of expert $i$ being selected, and $p_i$ denotes the average gating score of the expert $i$. Existing MoE training frameworks usually employ the parallel training strategy so that $f_i$ and the LBL are calculated within a $\\textbf{micro-batch}$ and then averaged across parallel groups. In essence, a mic"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.11873","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/2501.11873/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":"2501.11873","created_at":"2026-07-05T10:09:15.108878+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.11873v2","created_at":"2026-07-05T10:09:15.108878+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.11873","created_at":"2026-07-05T10:09:15.108878+00:00"},{"alias_kind":"pith_short_12","alias_value":"XCFF2J62UNVQ","created_at":"2026-07-05T10:09:15.108878+00:00"},{"alias_kind":"pith_short_16","alias_value":"XCFF2J62UNVQZ4GD","created_at":"2026-07-05T10:09:15.108878+00:00"},{"alias_kind":"pith_short_8","alias_value":"XCFF2J62","created_at":"2026-07-05T10:09:15.108878+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21228","citing_title":"Sakana Fugu Technical Report","ref_index":207,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08814","citing_title":"STAR: Rethinking MoE Routing as Structure-Aware Subspace Learning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01062","citing_title":"DAG-MoE: From Simple Mixture to Structural Aggregation in Mixture-of-Experts","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28835","citing_title":"Fisher-Routed Mixture of Experts for Federated Class-Incremental Learning","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31463","citing_title":"PithTrain: A Compact and Agent-Native MoE Training System","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08738","citing_title":"SlimQwen: Exploring the Pruning and Distillation in Large MoE Model Pre-training","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16690","citing_title":"UB-SMoE: Universally Balanced Sparse Mixture-of-Experts for Resource-adaptive Federated Fine-tuning of Foundation Models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2512.05564","citing_title":"ProPhy: Progressive Physical Alignment for Dynamic World Simulation","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2509.19349","citing_title":"ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution","ref_index":246,"is_internal_anchor":false},{"citing_arxiv_id":"2505.06708","citing_title":"Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08738","citing_title":"SlimQwen: Exploring the Pruning and Distillation in Large MoE Model Pre-training","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2505.09388","citing_title":"Qwen3 Technical Report","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD","json":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD.json","graph_json":"https://pith.science/api/pith-number/XCFF2J62UNVQZ4GDOD2EVQG2XD/graph.json","events_json":"https://pith.science/api/pith-number/XCFF2J62UNVQZ4GDOD2EVQG2XD/events.json","paper":"https://pith.science/paper/XCFF2J62"},"agent_actions":{"view_html":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD","download_json":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD.json","view_paper":"https://pith.science/paper/XCFF2J62","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.11873&json=true","fetch_graph":"https://pith.science/api/pith-number/XCFF2J62UNVQZ4GDOD2EVQG2XD/graph.json","fetch_events":"https://pith.science/api/pith-number/XCFF2J62UNVQZ4GDOD2EVQG2XD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD/action/storage_attestation","attest_author":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD/action/author_attestation","sign_citation":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD/action/citation_signature","submit_replication":"https://pith.science/pith/XCFF2J62UNVQZ4GDOD2EVQG2XD/action/replication_record"}},"created_at":"2026-07-05T10:09:15.108878+00:00","updated_at":"2026-07-05T10:09:15.108878+00:00"}