{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XRYAC2FZ2KBJWNF5EULNRJEOE5","short_pith_number":"pith:XRYAC2FZ","schema_version":"1.0","canonical_sha256":"bc700168b9d2829b34bd2516d8a48e2741814e97b2441b8486e0e60d109a8507","source":{"kind":"arxiv","id":"2602.09492","version":2},"attestation_state":"computed","paper":{"title":"Beware of the Batch Size: Hyperparameter Bias in Evaluating LoRA","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jaeho Lee, Sangyoon Lee","submitted_at":"2026-02-10T07:40:35Z","abstract_excerpt":"Low-rank adaptation (LoRA) is a standard approach for fine-tuning large language models, yet its many variants report conflicting empirical gains, often on the same benchmarks. We show that these contradictions arise from a single overlooked factor: the batch size. When properly tuned, vanilla LoRA often matches the performance of more complex variants. We further propose a proxy-based, cost-efficient strategy for batch size tuning, revealing the impact of rank, dataset size, and model capacity on the optimal batch size. Our findings elevate batch size from a minor implementation detail to a f"},"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":"2602.09492","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-02-10T07:40:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b6de80478111e4a8adc26608b37b8602fb438757f3204506a57dae126776c505","abstract_canon_sha256":"7e3749b80460b32d75a02bb723fc6e61097fc82d5631a231b65d8cd9e5e094a0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:51.222079Z","signature_b64":"78jNBKqU5JnajAuyDMGQXob/grVERPqmi7ijgQ1E8FINpzWPxN9mufGAzf+IX1HTW9jZk9T7Ezsncof3pHdwBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bc700168b9d2829b34bd2516d8a48e2741814e97b2441b8486e0e60d109a8507","last_reissued_at":"2026-06-02T02:04:51.221586Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:51.221586Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beware of the Batch Size: Hyperparameter Bias in Evaluating LoRA","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jaeho Lee, Sangyoon Lee","submitted_at":"2026-02-10T07:40:35Z","abstract_excerpt":"Low-rank adaptation (LoRA) is a standard approach for fine-tuning large language models, yet its many variants report conflicting empirical gains, often on the same benchmarks. We show that these contradictions arise from a single overlooked factor: the batch size. When properly tuned, vanilla LoRA often matches the performance of more complex variants. We further propose a proxy-based, cost-efficient strategy for batch size tuning, revealing the impact of rank, dataset size, and model capacity on the optimal batch size. Our findings elevate batch size from a minor implementation detail to a f"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.09492","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/2602.09492/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":"2602.09492","created_at":"2026-06-02T02:04:51.221645+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.09492v2","created_at":"2026-06-02T02:04:51.221645+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.09492","created_at":"2026-06-02T02:04:51.221645+00:00"},{"alias_kind":"pith_short_12","alias_value":"XRYAC2FZ2KBJ","created_at":"2026-06-02T02:04:51.221645+00:00"},{"alias_kind":"pith_short_16","alias_value":"XRYAC2FZ2KBJWNF5","created_at":"2026-06-02T02:04:51.221645+00:00"},{"alias_kind":"pith_short_8","alias_value":"XRYAC2FZ","created_at":"2026-06-02T02:04:51.221645+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2604.16332","citing_title":"Annotation Entropy Predicts Per-Example Learning Dynamics in LoRA Fine-Tuning","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2604.05426","citing_title":"ALTO: Adaptive LoRA Tuning and Orchestration for Heterogeneous LoRA Training Workloads","ref_index":62,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5","json":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5.json","graph_json":"https://pith.science/api/pith-number/XRYAC2FZ2KBJWNF5EULNRJEOE5/graph.json","events_json":"https://pith.science/api/pith-number/XRYAC2FZ2KBJWNF5EULNRJEOE5/events.json","paper":"https://pith.science/paper/XRYAC2FZ"},"agent_actions":{"view_html":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5","download_json":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5.json","view_paper":"https://pith.science/paper/XRYAC2FZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.09492&json=true","fetch_graph":"https://pith.science/api/pith-number/XRYAC2FZ2KBJWNF5EULNRJEOE5/graph.json","fetch_events":"https://pith.science/api/pith-number/XRYAC2FZ2KBJWNF5EULNRJEOE5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5/action/storage_attestation","attest_author":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5/action/author_attestation","sign_citation":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5/action/citation_signature","submit_replication":"https://pith.science/pith/XRYAC2FZ2KBJWNF5EULNRJEOE5/action/replication_record"}},"created_at":"2026-06-02T02:04:51.221645+00:00","updated_at":"2026-06-02T02:04:51.221645+00:00"}