{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:LG7OXAUU3ZVQSV3APGNC5WAJ63","short_pith_number":"pith:LG7OXAUU","schema_version":"1.0","canonical_sha256":"59beeb8294de6b095760799a2ed809f6f7d07d8af90d1456250d81e6c371e660","source":{"kind":"arxiv","id":"2407.17467","version":2},"attestation_state":"computed","paper":{"title":"CMR Scaling Law: Predicting Critical Mixture Ratios for Continual Pre-training of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chuanghao Ding, Fei Tan, Jiawei Gu, Rui Zhao, Zacc Yang","submitted_at":"2024-07-24T17:59:02Z","abstract_excerpt":"Large Language Models (LLMs) excel in diverse tasks but often underperform in specialized fields due to limited domain-specific or proprietary corpus. Continual pre-training (CPT) enhances LLM capabilities by imbuing new domain-specific or proprietary knowledge while replaying general corpus to prevent catastrophic forgetting. The data mixture ratio of general corpus and domain-specific corpus, however, has been chosen heuristically, leading to sub-optimal training efficiency in practice. In this context, we attempt to re-visit the scaling behavior of LLMs under the hood of CPT, and discover a"},"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":"2407.17467","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-07-24T17:59:02Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e0f0e29612fe5f86090864b3f846c26afda36196d058751d48b9f28a9dbc9c9a","abstract_canon_sha256":"9542e36b90503706846624e34ed9b5321862547854b9980ee02ee2aeb0317f96"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:16:51.638925Z","signature_b64":"Bdwu3lNfgwx81wYOj65rpmhzH7KXHlELu2n+ELNRDdcP6EbHAx+JjXsQL+jLIMb0E8Hk+5UHdZ6i/cFjwPOYAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"59beeb8294de6b095760799a2ed809f6f7d07d8af90d1456250d81e6c371e660","last_reissued_at":"2026-07-05T09:16:51.638296Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:16:51.638296Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CMR Scaling Law: Predicting Critical Mixture Ratios for Continual Pre-training of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chuanghao Ding, Fei Tan, Jiawei Gu, Rui Zhao, Zacc Yang","submitted_at":"2024-07-24T17:59:02Z","abstract_excerpt":"Large Language Models (LLMs) excel in diverse tasks but often underperform in specialized fields due to limited domain-specific or proprietary corpus. Continual pre-training (CPT) enhances LLM capabilities by imbuing new domain-specific or proprietary knowledge while replaying general corpus to prevent catastrophic forgetting. The data mixture ratio of general corpus and domain-specific corpus, however, has been chosen heuristically, leading to sub-optimal training efficiency in practice. In this context, we attempt to re-visit the scaling behavior of LLMs under the hood of CPT, and discover a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.17467","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/2407.17467/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":"2407.17467","created_at":"2026-07-05T09:16:51.638373+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.17467v2","created_at":"2026-07-05T09:16:51.638373+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.17467","created_at":"2026-07-05T09:16:51.638373+00:00"},{"alias_kind":"pith_short_12","alias_value":"LG7OXAUU3ZVQ","created_at":"2026-07-05T09:16:51.638373+00:00"},{"alias_kind":"pith_short_16","alias_value":"LG7OXAUU3ZVQSV3A","created_at":"2026-07-05T09:16:51.638373+00:00"},{"alias_kind":"pith_short_8","alias_value":"LG7OXAUU","created_at":"2026-07-05T09:16:51.638373+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2409.06624","citing_title":"A Practice of Post-Training on Llama-3 70B with Optimal Selection of Additional Language Mixture Ratio","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15104","citing_title":"From Text to Voice: A Reproducible and Verifiable Framework for Evaluating Tool Calling LLM Agents","ref_index":169,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63","json":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63.json","graph_json":"https://pith.science/api/pith-number/LG7OXAUU3ZVQSV3APGNC5WAJ63/graph.json","events_json":"https://pith.science/api/pith-number/LG7OXAUU3ZVQSV3APGNC5WAJ63/events.json","paper":"https://pith.science/paper/LG7OXAUU"},"agent_actions":{"view_html":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63","download_json":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63.json","view_paper":"https://pith.science/paper/LG7OXAUU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.17467&json=true","fetch_graph":"https://pith.science/api/pith-number/LG7OXAUU3ZVQSV3APGNC5WAJ63/graph.json","fetch_events":"https://pith.science/api/pith-number/LG7OXAUU3ZVQSV3APGNC5WAJ63/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63/action/storage_attestation","attest_author":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63/action/author_attestation","sign_citation":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63/action/citation_signature","submit_replication":"https://pith.science/pith/LG7OXAUU3ZVQSV3APGNC5WAJ63/action/replication_record"}},"created_at":"2026-07-05T09:16:51.638373+00:00","updated_at":"2026-07-05T09:16:51.638373+00:00"}