{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BGXKFVRSLUB5AZY5LPJULJB4LN","short_pith_number":"pith:BGXKFVRS","schema_version":"1.0","canonical_sha256":"09aea2d6325d03d0671d5bd345a43c5b65288d54994c40de8967264a9dc53967","source":{"kind":"arxiv","id":"2405.03548","version":4},"attestation_state":"computed","paper":{"title":"MAmmoTH2: Scaling Instructions from the Web","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ge Zhang, Tuney Zheng, Wenhu Chen, Xiang Yue","submitted_at":"2024-05-06T15:11:38Z","abstract_excerpt":"Instruction tuning improves the reasoning abilities of large language models (LLMs), with data quality and scalability being the crucial factors. Most instruction tuning data come from human crowd-sourcing or GPT-4 distillation. We propose a paradigm to efficiently harvest 10 million naturally existing instruction data from the pre-training web corpus to enhance LLM reasoning. Our approach involves (1) recalling relevant documents, (2) extracting instruction-response pairs, and (3) refining the extracted pairs using open-source LLMs. Fine-tuning base LLMs on this dataset, we build MAmmoTH2 mod"},"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":"2405.03548","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-05-06T15:11:38Z","cross_cats_sorted":[],"title_canon_sha256":"c3f706a8f256b6f4afcb0e8131f83d9341178196648e7a30e9c92c1df93166fc","abstract_canon_sha256":"27bf74798f1db4032b52c25c5490598a57d5b6652641d864f5a2d28b9f33f788"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:22:23.091574Z","signature_b64":"N6VT5yrmdBk11A+G//kk1QkDvt3jeL78ywTAOfv5g4uQVmtBY6d/fbHizp5fN7BES5j1NrKZjk9CeY1k+y5dCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"09aea2d6325d03d0671d5bd345a43c5b65288d54994c40de8967264a9dc53967","last_reissued_at":"2026-07-05T08:22:23.091105Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:22:23.091105Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MAmmoTH2: Scaling Instructions from the Web","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ge Zhang, Tuney Zheng, Wenhu Chen, Xiang Yue","submitted_at":"2024-05-06T15:11:38Z","abstract_excerpt":"Instruction tuning improves the reasoning abilities of large language models (LLMs), with data quality and scalability being the crucial factors. Most instruction tuning data come from human crowd-sourcing or GPT-4 distillation. We propose a paradigm to efficiently harvest 10 million naturally existing instruction data from the pre-training web corpus to enhance LLM reasoning. Our approach involves (1) recalling relevant documents, (2) extracting instruction-response pairs, and (3) refining the extracted pairs using open-source LLMs. Fine-tuning base LLMs on this dataset, we build MAmmoTH2 mod"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.03548","kind":"arxiv","version":4},"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/2405.03548/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":"2405.03548","created_at":"2026-07-05T08:22:23.091164+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.03548v4","created_at":"2026-07-05T08:22:23.091164+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.03548","created_at":"2026-07-05T08:22:23.091164+00:00"},{"alias_kind":"pith_short_12","alias_value":"BGXKFVRSLUB5","created_at":"2026-07-05T08:22:23.091164+00:00"},{"alias_kind":"pith_short_16","alias_value":"BGXKFVRSLUB5AZY5","created_at":"2026-07-05T08:22:23.091164+00:00"},{"alias_kind":"pith_short_8","alias_value":"BGXKFVRS","created_at":"2026-07-05T08:22:23.091164+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2502.02871","citing_title":"Position: Multimodal Large Language Models Can Significantly Advance Scientific Reasoning","ref_index":243,"is_internal_anchor":false},{"citing_arxiv_id":"2406.18629","citing_title":"Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2406.11794","citing_title":"DataComp-LM: In search of the next generation of training sets for language models","ref_index":210,"is_internal_anchor":false},{"citing_arxiv_id":"2502.01456","citing_title":"Process Reinforcement through Implicit Rewards","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2406.01574","citing_title":"MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10079","citing_title":"Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN","json":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN.json","graph_json":"https://pith.science/api/pith-number/BGXKFVRSLUB5AZY5LPJULJB4LN/graph.json","events_json":"https://pith.science/api/pith-number/BGXKFVRSLUB5AZY5LPJULJB4LN/events.json","paper":"https://pith.science/paper/BGXKFVRS"},"agent_actions":{"view_html":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN","download_json":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN.json","view_paper":"https://pith.science/paper/BGXKFVRS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.03548&json=true","fetch_graph":"https://pith.science/api/pith-number/BGXKFVRSLUB5AZY5LPJULJB4LN/graph.json","fetch_events":"https://pith.science/api/pith-number/BGXKFVRSLUB5AZY5LPJULJB4LN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN/action/storage_attestation","attest_author":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN/action/author_attestation","sign_citation":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN/action/citation_signature","submit_replication":"https://pith.science/pith/BGXKFVRSLUB5AZY5LPJULJB4LN/action/replication_record"}},"created_at":"2026-07-05T08:22:23.091164+00:00","updated_at":"2026-07-05T08:22:23.091164+00:00"}