{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CXKT2WJBWUQKD73OXIAF2RAG4T","short_pith_number":"pith:CXKT2WJB","schema_version":"1.0","canonical_sha256":"15d53d5921b520a1ff6eba005d4406e4d077f2cd85356432ed1dbe236f630404","source":{"kind":"arxiv","id":"2404.00934","version":2},"attestation_state":"computed","paper":{"title":"ChatGLM-RLHF: Practices of Aligning Large Language Models with Human Feedback","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Aohan Zeng, Hongning Wang, Jie Tang, Minlie Huang, Qinkai Zheng, Xiaohan Zhang, Xiao Liu, Yilin Niu, Yuxiao Dong, Zhengxiao Du, Zhenyu Hou","submitted_at":"2024-04-01T05:39:36Z","abstract_excerpt":"ChatGLM is a free-to-use AI service powered by the ChatGLM family of large language models (LLMs). In this paper, we present the ChatGLM-RLHF pipeline -- a reinforcement learning from human feedback (RLHF) system -- designed to enhance ChatGLM's alignment with human preferences. ChatGLM-RLHF encompasses three major components: the collection of human preference data, the training of the reward model, and the optimization of policies. Throughout the process of integrating ChatGLM-RLHF into production, we encountered and addressed several unprecedented challenges. We introduce the strategies to "},"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":"2404.00934","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-04-01T05:39:36Z","cross_cats_sorted":[],"title_canon_sha256":"13f2a7abb05498d37aeaefcc9a938b0b0d36096e1fbb3add25cd9876143f7f08","abstract_canon_sha256":"e1dfde1e930e245983de2ebb3aca1e88f2fab98e1478d36e091a3bba3ac4192d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:03:51.483222Z","signature_b64":"Pd5m+9poAbIpwOBeQEvaVZBwSiwmoMjDUVYYr8hilr/VCSfyISxmW/IOF88aVbcu6xcNr0DgLJfiXmZjkUqrAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"15d53d5921b520a1ff6eba005d4406e4d077f2cd85356432ed1dbe236f630404","last_reissued_at":"2026-07-05T08:03:51.482784Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:03:51.482784Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ChatGLM-RLHF: Practices of Aligning Large Language Models with Human Feedback","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Aohan Zeng, Hongning Wang, Jie Tang, Minlie Huang, Qinkai Zheng, Xiaohan Zhang, Xiao Liu, Yilin Niu, Yuxiao Dong, Zhengxiao Du, Zhenyu Hou","submitted_at":"2024-04-01T05:39:36Z","abstract_excerpt":"ChatGLM is a free-to-use AI service powered by the ChatGLM family of large language models (LLMs). In this paper, we present the ChatGLM-RLHF pipeline -- a reinforcement learning from human feedback (RLHF) system -- designed to enhance ChatGLM's alignment with human preferences. ChatGLM-RLHF encompasses three major components: the collection of human preference data, the training of the reward model, and the optimization of policies. Throughout the process of integrating ChatGLM-RLHF into production, we encountered and addressed several unprecedented challenges. We introduce the strategies to "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.00934","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/2404.00934/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":"2404.00934","created_at":"2026-07-05T08:03:51.482844+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.00934v2","created_at":"2026-07-05T08:03:51.482844+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.00934","created_at":"2026-07-05T08:03:51.482844+00:00"},{"alias_kind":"pith_short_12","alias_value":"CXKT2WJBWUQK","created_at":"2026-07-05T08:03:51.482844+00:00"},{"alias_kind":"pith_short_16","alias_value":"CXKT2WJBWUQKD73O","created_at":"2026-07-05T08:03:51.482844+00:00"},{"alias_kind":"pith_short_8","alias_value":"CXKT2WJB","created_at":"2026-07-05T08:03:51.482844+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2601.21484","citing_title":"ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2506.18841","citing_title":"LongWriter-Zero: Mastering Ultra-Long Text Generation via Reinforcement Learning","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2510.20129","citing_title":"SAID: Safety-Aware Intent Defense via Prefix Probing for Large Language Models","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2511.01014","citing_title":"IF-CRITIC: Towards a Fine-Grained LLM Critic for Instruction-Following Evaluation","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2601.21484","citing_title":"ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T","json":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T.json","graph_json":"https://pith.science/api/pith-number/CXKT2WJBWUQKD73OXIAF2RAG4T/graph.json","events_json":"https://pith.science/api/pith-number/CXKT2WJBWUQKD73OXIAF2RAG4T/events.json","paper":"https://pith.science/paper/CXKT2WJB"},"agent_actions":{"view_html":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T","download_json":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T.json","view_paper":"https://pith.science/paper/CXKT2WJB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.00934&json=true","fetch_graph":"https://pith.science/api/pith-number/CXKT2WJBWUQKD73OXIAF2RAG4T/graph.json","fetch_events":"https://pith.science/api/pith-number/CXKT2WJBWUQKD73OXIAF2RAG4T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T/action/storage_attestation","attest_author":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T/action/author_attestation","sign_citation":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T/action/citation_signature","submit_replication":"https://pith.science/pith/CXKT2WJBWUQKD73OXIAF2RAG4T/action/replication_record"}},"created_at":"2026-07-05T08:03:51.482844+00:00","updated_at":"2026-07-05T08:03:51.482844+00:00"}