{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:AUZ2KXEGQ3FFCUNGONF2K3MBBM","short_pith_number":"pith:AUZ2KXEG","schema_version":"1.0","canonical_sha256":"0533a55c8686ca5151a6734ba56d810b2eeb6add69cb357379b26a53780161e5","source":{"kind":"arxiv","id":"2412.08347","version":1},"attestation_state":"computed","paper":{"title":"SmolTulu: Higher Learning Rate to Batch Size Ratios Can Lead to Better Reasoning in SLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Sultan Alrashed","submitted_at":"2024-12-11T12:41:36Z","abstract_excerpt":"We present SmolTulu-1.7b-Instruct, referenced in this report as SmolTulu-DPO-1130, an instruction-tuned language model that adapts AllenAI's Tulu 3 post-training pipeline to enhance Huggingface's SmolLM2-1.7B base model. Through comprehensive empirical analysis using a 135M parameter model, we demonstrate that the relationship between learning rate and batch size significantly impacts model performance in a task-dependent manner. Our findings reveal a clear split: reasoning tasks like ARC and GSM8K benefit from higher learning rate to batch size ratios, while pattern recognition tasks such as "},"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":"2412.08347","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-11T12:41:36Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2a4eff7892fd81983f732ee978394e6f2775bdd4f0a5802ac013190a0dfbcb1d","abstract_canon_sha256":"ae66babe75a7f36fe26a06b25e569789c245426c6aced5ff563877cfd7a91bf2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:47:48.508765Z","signature_b64":"qwWuN4kR7EJAwIZaQXlg7o3/uxdDqAoOelxH26J6ur/sOwAHZOLvDjfvOP2xa3UiT8vF3PjJt/DgCeaUFQEnBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0533a55c8686ca5151a6734ba56d810b2eeb6add69cb357379b26a53780161e5","last_reissued_at":"2026-07-05T09:47:48.508370Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:47:48.508370Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SmolTulu: Higher Learning Rate to Batch Size Ratios Can Lead to Better Reasoning in SLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Sultan Alrashed","submitted_at":"2024-12-11T12:41:36Z","abstract_excerpt":"We present SmolTulu-1.7b-Instruct, referenced in this report as SmolTulu-DPO-1130, an instruction-tuned language model that adapts AllenAI's Tulu 3 post-training pipeline to enhance Huggingface's SmolLM2-1.7B base model. Through comprehensive empirical analysis using a 135M parameter model, we demonstrate that the relationship between learning rate and batch size significantly impacts model performance in a task-dependent manner. Our findings reveal a clear split: reasoning tasks like ARC and GSM8K benefit from higher learning rate to batch size ratios, while pattern recognition tasks such as "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.08347","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/2412.08347/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":"2412.08347","created_at":"2026-07-05T09:47:48.508426+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.08347v1","created_at":"2026-07-05T09:47:48.508426+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.08347","created_at":"2026-07-05T09:47:48.508426+00:00"},{"alias_kind":"pith_short_12","alias_value":"AUZ2KXEGQ3FF","created_at":"2026-07-05T09:47:48.508426+00:00"},{"alias_kind":"pith_short_16","alias_value":"AUZ2KXEGQ3FFCUNG","created_at":"2026-07-05T09:47:48.508426+00:00"},{"alias_kind":"pith_short_8","alias_value":"AUZ2KXEG","created_at":"2026-07-05T09:47:48.508426+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2504.12501","citing_title":"Reinforcement Learning from Human Feedback","ref_index":59,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM","json":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM.json","graph_json":"https://pith.science/api/pith-number/AUZ2KXEGQ3FFCUNGONF2K3MBBM/graph.json","events_json":"https://pith.science/api/pith-number/AUZ2KXEGQ3FFCUNGONF2K3MBBM/events.json","paper":"https://pith.science/paper/AUZ2KXEG"},"agent_actions":{"view_html":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM","download_json":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM.json","view_paper":"https://pith.science/paper/AUZ2KXEG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.08347&json=true","fetch_graph":"https://pith.science/api/pith-number/AUZ2KXEGQ3FFCUNGONF2K3MBBM/graph.json","fetch_events":"https://pith.science/api/pith-number/AUZ2KXEGQ3FFCUNGONF2K3MBBM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM/action/storage_attestation","attest_author":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM/action/author_attestation","sign_citation":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM/action/citation_signature","submit_replication":"https://pith.science/pith/AUZ2KXEGQ3FFCUNGONF2K3MBBM/action/replication_record"}},"created_at":"2026-07-05T09:47:48.508426+00:00","updated_at":"2026-07-05T09:47:48.508426+00:00"}