{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:AB76BDVVRSXTKIYKOGZVFXJRVS","short_pith_number":"pith:AB76BDVV","schema_version":"1.0","canonical_sha256":"007fe08eb58caf35230a71b352dd31ac83fc8a4f594f087adb79fc6b2085dd1b","source":{"kind":"arxiv","id":"2208.03306","version":1},"attestation_state":"computed","paper":{"title":"Branch-Train-Merge: Embarrassingly Parallel Training of Expert Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Luke Zettlemoyer, Margaret Li, Mike Lewis, Noah A. Smith, Suchin Gururangan, Tim Althoff, Tim Dettmers","submitted_at":"2022-08-05T17:46:38Z","abstract_excerpt":"We present Branch-Train-Merge (BTM), a communication-efficient algorithm for embarrassingly parallel training of large language models (LLMs). We show it is possible to independently train subparts of a new class of LLMs on different subsets of the data, eliminating the massive multi-node synchronization currently required to train LLMs. BTM learns a set of independent expert LMs (ELMs), each specialized to a different textual domain, such as scientific or legal text. These ELMs can be added and removed to update data coverage, ensembled to generalize to new domains, or averaged to collapse ba"},"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":"2208.03306","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-08-05T17:46:38Z","cross_cats_sorted":[],"title_canon_sha256":"153366eabcca4bd87da29b01a906fa6a7fdb4e0148ad104e618fcfa85d24ff1e","abstract_canon_sha256":"04b97c2e77c8efcf9c6ba8ede1a4afcfc4230e4cf0057dbac9ceeb3929850513"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:46:22.903423Z","signature_b64":"YYl+sUZjVMTRl1MVT51vUtfjsvmjQaRm3P7cndAMgyTaFbYIFVI+L+r1mMoUkvt+183NxArvL+iGiIusvHm9Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"007fe08eb58caf35230a71b352dd31ac83fc8a4f594f087adb79fc6b2085dd1b","last_reissued_at":"2026-07-05T04:46:22.902937Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:46:22.902937Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Branch-Train-Merge: Embarrassingly Parallel Training of Expert Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Luke Zettlemoyer, Margaret Li, Mike Lewis, Noah A. Smith, Suchin Gururangan, Tim Althoff, Tim Dettmers","submitted_at":"2022-08-05T17:46:38Z","abstract_excerpt":"We present Branch-Train-Merge (BTM), a communication-efficient algorithm for embarrassingly parallel training of large language models (LLMs). We show it is possible to independently train subparts of a new class of LLMs on different subsets of the data, eliminating the massive multi-node synchronization currently required to train LLMs. BTM learns a set of independent expert LMs (ELMs), each specialized to a different textual domain, such as scientific or legal text. These ELMs can be added and removed to update data coverage, ensembled to generalize to new domains, or averaged to collapse ba"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2208.03306","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/2208.03306/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":"2208.03306","created_at":"2026-07-05T04:46:22.902988+00:00"},{"alias_kind":"arxiv_version","alias_value":"2208.03306v1","created_at":"2026-07-05T04:46:22.902988+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2208.03306","created_at":"2026-07-05T04:46:22.902988+00:00"},{"alias_kind":"pith_short_12","alias_value":"AB76BDVVRSXT","created_at":"2026-07-05T04:46:22.902988+00:00"},{"alias_kind":"pith_short_16","alias_value":"AB76BDVVRSXTKIYK","created_at":"2026-07-05T04:46:22.902988+00:00"},{"alias_kind":"pith_short_8","alias_value":"AB76BDVV","created_at":"2026-07-05T04:46:22.902988+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24722","citing_title":"Decentralised AI Training and Inference with BlockTrain","ref_index":101,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00620","citing_title":"Identifying Latent Concepts and Structures for Generalized Category Discovery","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31796","citing_title":"CHERRY: Compressed Hierarchical Experts with Recurrent Representational Yield","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2408.01119","citing_title":"Task Prompt Vectors: Effective Initialization through Multi-Task Soft-Prompt Transfer","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2504.12501","citing_title":"Reinforcement Learning from Human Feedback","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19095","citing_title":"ScheduleFree+: Scaling Learning-Rate-Free & Schedule-Free Learning to Large Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2408.07666","citing_title":"Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities","ref_index":121,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13997","citing_title":"HodgeCover: Higher-Order Topological Coverage Drives Compression of Sparse Mixture-of-Experts","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14289","citing_title":"MetaMoE: Diversity-Aware Proxy Selection for Privacy-Preserving Mixture-of-Experts Unification","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02719","citing_title":"MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applications","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2212.04089","citing_title":"Editing Models with Task Arithmetic","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11170","citing_title":"Unlearning with Asymmetric Sources: Improved Unlearning-Utility Trade-off with Public Data","ref_index":162,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18473","citing_title":"Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS","json":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS.json","graph_json":"https://pith.science/api/pith-number/AB76BDVVRSXTKIYKOGZVFXJRVS/graph.json","events_json":"https://pith.science/api/pith-number/AB76BDVVRSXTKIYKOGZVFXJRVS/events.json","paper":"https://pith.science/paper/AB76BDVV"},"agent_actions":{"view_html":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS","download_json":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS.json","view_paper":"https://pith.science/paper/AB76BDVV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2208.03306&json=true","fetch_graph":"https://pith.science/api/pith-number/AB76BDVVRSXTKIYKOGZVFXJRVS/graph.json","fetch_events":"https://pith.science/api/pith-number/AB76BDVVRSXTKIYKOGZVFXJRVS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS/action/storage_attestation","attest_author":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS/action/author_attestation","sign_citation":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS/action/citation_signature","submit_replication":"https://pith.science/pith/AB76BDVVRSXTKIYKOGZVFXJRVS/action/replication_record"}},"created_at":"2026-07-05T04:46:22.902988+00:00","updated_at":"2026-07-05T04:46:22.902988+00:00"}