{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:TGZJHWJEIAJ77JAEUKZKFENVYZ","short_pith_number":"pith:TGZJHWJE","canonical_record":{"source":{"id":"2410.18451","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-10-24T06:06:26Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"ba1d741daa5ae514a398bd366ef016d9f380d167ec248825d5a97bd0128ca95b","abstract_canon_sha256":"a75a2a37f76b929d023147fbbb3051e40e795a8c5e309798c29d5d682419a308"},"schema_version":"1.0"},"canonical_sha256":"99b293d9244013ffa404a2b2a291b5c65de3c0fccc9cb70b1b33d7e1e915fff5","source":{"kind":"arxiv","id":"2410.18451","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2410.18451","created_at":"2026-05-17T23:38:13Z"},{"alias_kind":"arxiv_version","alias_value":"2410.18451v1","created_at":"2026-05-17T23:38:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.18451","created_at":"2026-05-17T23:38:13Z"},{"alias_kind":"pith_short_12","alias_value":"TGZJHWJEIAJ7","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"TGZJHWJEIAJ77JAE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"TGZJHWJE","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:TGZJHWJEIAJ77JAEUKZKFENVYZ","target":"record","payload":{"canonical_record":{"source":{"id":"2410.18451","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-10-24T06:06:26Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"ba1d741daa5ae514a398bd366ef016d9f380d167ec248825d5a97bd0128ca95b","abstract_canon_sha256":"a75a2a37f76b929d023147fbbb3051e40e795a8c5e309798c29d5d682419a308"},"schema_version":"1.0"},"canonical_sha256":"99b293d9244013ffa404a2b2a291b5c65de3c0fccc9cb70b1b33d7e1e915fff5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:13.611885Z","signature_b64":"0amWQXZfK3DBC2EZ113WmxhWcbOpw+KocGzzr2tZgynbvSEpIfFtZZC0cMOJF9kVxxd0G6rHr9sw2XUmnK0lBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"99b293d9244013ffa404a2b2a291b5c65de3c0fccc9cb70b1b33d7e1e915fff5","last_reissued_at":"2026-05-17T23:38:13.611332Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:13.611332Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2410.18451","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:13Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"qOJG4HdfkpT0xMHNI5+7zsrT9VMjQPTs6vxEGDOcPvKBUJDHy09xLwWp8trRj1IdgcZZVlffmPGCCDjuKNhGDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-09T07:52:16.028302Z"},"content_sha256":"4a111a3ec4ccb6b9fba4534fdc551d0e127d9a5eca570a222a1f52ac084ba6c5","schema_version":"1.0","event_id":"sha256:4a111a3ec4ccb6b9fba4534fdc551d0e127d9a5eca570a222a1f52ac084ba6c5"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:TGZJHWJEIAJ77JAEUKZKFENVYZ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Strategic data selection and filtering from open-source pairs yields top-ranked reward models with just 80K examples.","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Chaojie Wang, Chris Yuhao Liu, Jiacai Liu, Jujie He, Liang Zeng, Rui Yan, Shuicheng Yan, Yahui Zhou, Yang Liu","submitted_at":"2024-10-24T06:06:26Z","abstract_excerpt":"In this report, we introduce a collection of methods to enhance reward modeling for LLMs, focusing specifically on data-centric techniques. We propose effective data selection and filtering strategies for curating high-quality open-source preference datasets, culminating in the Skywork-Reward data collection, which contains only 80K preference pairs -- significantly smaller than existing datasets. Using this curated dataset, we developed the Skywork-Reward model series -- Skywork-Reward-Gemma-27B and Skywork-Reward-Llama-3.1-8B -- with the former currently holding the top position on the Rewar"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"our techniques and datasets have directly enhanced the performance of many top-ranked models on RewardBench","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the data selection and filtering strategies produce generalizable improvements rather than leaderboard-specific gains tied to the particular open-source sources and evaluation distribution.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Strategic data selection and filtering from open-source pairs yields top-ranked reward models with just 80K examples.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"780d81392f9513f91bf1727b6cd043af459d887dab08533f8aec86585c574627"},"source":{"id":"2410.18451","kind":"arxiv","version":1},"verdict":{"id":"0c66f76f-6b15-4dd0-9903-038264b491f4","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T16:13:10.262455Z","strongest_claim":"our techniques and datasets have directly enhanced the performance of many top-ranked models on RewardBench","one_line_summary":"Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the data selection and filtering strategies produce generalizable improvements rather than leaderboard-specific gains tied to the particular open-source sources and evaluation distribution.","pith_extraction_headline":"Strategic data selection and filtering from open-source pairs yields top-ranked reward models with just 80K examples."},"references":{"count":27,"sample":[{"doi":"","year":null,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","ref_index":1,"cited_arxiv_id":"2303.08774","is_internal_anchor":true},{"doi":"","year":null,"title":"Nemotron-4 340b technical report","work_id":"f462f380-6529-48f1-a3ff-1c847606a77e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","ref_index":3,"cited_arxiv_id":"2204.05862","is_internal_anchor":true},{"doi":"","year":null,"title":"M. Bellagente, J. Tow, D. Mahan, D. Phung, M. Zhuravinskyi, R. Adithyan, J. Baicoianu, B. Brooks, N. Cooper, A. Datta, et al. Stable lm 2 1.6 b technical report. arXiv preprint arXiv:2402.17834,","work_id":"5ed7b6c0-8393-4f50-a699-44ae533adc22","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"InternLM2 Technical Report","work_id":"dfa13e0e-1c3c-4fb6-943d-a19945bacdbe","ref_index":5,"cited_arxiv_id":"2403.17297","is_internal_anchor":true}],"resolved_work":27,"snapshot_sha256":"f7484d79d6da4d0fcb61f8b3a11e465fe8d7775480760366ddfb929ec3e59bfe","internal_anchors":13},"formal_canon":{"evidence_count":3,"snapshot_sha256":"1f60d6648a19a371eb4e9ec9907ce8ea403bb920f68d0ea07f8b785fa6699abb"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"0c66f76f-6b15-4dd0-9903-038264b491f4"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:13Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"pCN/qUsZsPQEaBEbs+qhPfYjpK42xDmL670NS6jYNrVq11OsIFo1uFvm4JcOmIZja0ZJAZRjh+cOoaTj8PC9DQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-09T07:52:16.029341Z"},"content_sha256":"63dfc48e75ef5ecd749a1ba925d47a3a56e3f4c0b404426a66f2ad69cd1ec0c5","schema_version":"1.0","event_id":"sha256:63dfc48e75ef5ecd749a1ba925d47a3a56e3f4c0b404426a66f2ad69cd1ec0c5"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ/bundle.json","state_url":"https://pith.science/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-09T07:52:16Z","links":{"resolver":"https://pith.science/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ","bundle":"https://pith.science/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ/bundle.json","state":"https://pith.science/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/TGZJHWJEIAJ77JAEUKZKFENVYZ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:TGZJHWJEIAJ77JAEUKZKFENVYZ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"a75a2a37f76b929d023147fbbb3051e40e795a8c5e309798c29d5d682419a308","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-10-24T06:06:26Z","title_canon_sha256":"ba1d741daa5ae514a398bd366ef016d9f380d167ec248825d5a97bd0128ca95b"},"schema_version":"1.0","source":{"id":"2410.18451","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2410.18451","created_at":"2026-05-17T23:38:13Z"},{"alias_kind":"arxiv_version","alias_value":"2410.18451v1","created_at":"2026-05-17T23:38:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.18451","created_at":"2026-05-17T23:38:13Z"},{"alias_kind":"pith_short_12","alias_value":"TGZJHWJEIAJ7","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"TGZJHWJEIAJ77JAE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"TGZJHWJE","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:63dfc48e75ef5ecd749a1ba925d47a3a56e3f4c0b404426a66f2ad69cd1ec0c5","target":"graph","created_at":"2026-05-17T23:38:13Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"our techniques and datasets have directly enhanced the performance of many top-ranked models on RewardBench"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the data selection and filtering strategies produce generalizable improvements rather than leaderboard-specific gains tied to the particular open-source sources and evaluation distribution."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Strategic data selection and filtering from open-source pairs yields top-ranked reward models with just 80K examples."}],"snapshot_sha256":"780d81392f9513f91bf1727b6cd043af459d887dab08533f8aec86585c574627"},"formal_canon":{"evidence_count":3,"snapshot_sha256":"1f60d6648a19a371eb4e9ec9907ce8ea403bb920f68d0ea07f8b785fa6699abb"},"paper":{"abstract_excerpt":"In this report, we introduce a collection of methods to enhance reward modeling for LLMs, focusing specifically on data-centric techniques. We propose effective data selection and filtering strategies for curating high-quality open-source preference datasets, culminating in the Skywork-Reward data collection, which contains only 80K preference pairs -- significantly smaller than existing datasets. Using this curated dataset, we developed the Skywork-Reward model series -- Skywork-Reward-Gemma-27B and Skywork-Reward-Llama-3.1-8B -- with the former currently holding the top position on the Rewar","authors_text":"Chaojie Wang, Chris Yuhao Liu, Jiacai Liu, Jujie He, Liang Zeng, Rui Yan, Shuicheng Yan, Yahui Zhou, Yang Liu","cross_cats":["cs.CL"],"headline":"Strategic data selection and filtering from open-source pairs yields top-ranked reward models with just 80K examples.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-10-24T06:06:26Z","title":"Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs"},"references":{"count":27,"internal_anchors":13,"resolved_work":27,"sample":[{"cited_arxiv_id":"2303.08774","doi":"","is_internal_anchor":true,"ref_index":1,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Nemotron-4 340b technical report","work_id":"f462f380-6529-48f1-a3ff-1c847606a77e","year":null},{"cited_arxiv_id":"2204.05862","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"M. Bellagente, J. Tow, D. Mahan, D. Phung, M. Zhuravinskyi, R. Adithyan, J. Baicoianu, B. Brooks, N. Cooper, A. Datta, et al. Stable lm 2 1.6 b technical report. arXiv preprint arXiv:2402.17834,","work_id":"5ed7b6c0-8393-4f50-a699-44ae533adc22","year":null},{"cited_arxiv_id":"2403.17297","doi":"","is_internal_anchor":true,"ref_index":5,"title":"InternLM2 Technical Report","work_id":"dfa13e0e-1c3c-4fb6-943d-a19945bacdbe","year":null}],"snapshot_sha256":"f7484d79d6da4d0fcb61f8b3a11e465fe8d7775480760366ddfb929ec3e59bfe"},"source":{"id":"2410.18451","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-17T16:13:10.262455Z","id":"0c66f76f-6b15-4dd0-9903-038264b491f4","model_set":{"reader":"grok-4.3"},"one_line_summary":"Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Strategic data selection and filtering from open-source pairs yields top-ranked reward models with just 80K examples.","strongest_claim":"our techniques and datasets have directly enhanced the performance of many top-ranked models on RewardBench","weakest_assumption":"That the data selection and filtering strategies produce generalizable improvements rather than leaderboard-specific gains tied to the particular open-source sources and evaluation distribution."}},"verdict_id":"0c66f76f-6b15-4dd0-9903-038264b491f4"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:4a111a3ec4ccb6b9fba4534fdc551d0e127d9a5eca570a222a1f52ac084ba6c5","target":"record","created_at":"2026-05-17T23:38:13Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"a75a2a37f76b929d023147fbbb3051e40e795a8c5e309798c29d5d682419a308","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-10-24T06:06:26Z","title_canon_sha256":"ba1d741daa5ae514a398bd366ef016d9f380d167ec248825d5a97bd0128ca95b"},"schema_version":"1.0","source":{"id":"2410.18451","kind":"arxiv","version":1}},"canonical_sha256":"99b293d9244013ffa404a2b2a291b5c65de3c0fccc9cb70b1b33d7e1e915fff5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"99b293d9244013ffa404a2b2a291b5c65de3c0fccc9cb70b1b33d7e1e915fff5","first_computed_at":"2026-05-17T23:38:13.611332Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:13.611332Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"0amWQXZfK3DBC2EZ113WmxhWcbOpw+KocGzzr2tZgynbvSEpIfFtZZC0cMOJF9kVxxd0G6rHr9sw2XUmnK0lBw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:13.611885Z","signed_message":"canonical_sha256_bytes"},"source_id":"2410.18451","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:4a111a3ec4ccb6b9fba4534fdc551d0e127d9a5eca570a222a1f52ac084ba6c5","sha256:63dfc48e75ef5ecd749a1ba925d47a3a56e3f4c0b404426a66f2ad69cd1ec0c5"],"state_sha256":"4d08d83269e2c80b8018b5bd683305083db809417e7e89fc6d2c83375ad60137"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Hr+gcvt1Ozk0CNIP+0JuI8X5/OeUBI00JICB7Pb3t36tO0w/y4BSSTPBxfRx6tnkJC4OqQG9dFmXErqLmKhaAA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-09T07:52:16.034170Z","bundle_sha256":"b9ad13233979ebf9b68f91fbd203f1bdaa0a189955a23d7b3b0d77e9e8d17f7a"}}