{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5PU2B5CYD36UGQXFRFF2AK44SX","short_pith_number":"pith:5PU2B5CY","schema_version":"1.0","canonical_sha256":"ebe9a0f4581efd4342e5894ba02b9c95e76f59c96d3a5b4dd24bcfe05808b8ec","source":{"kind":"arxiv","id":"2502.18407","version":1},"attestation_state":"computed","paper":{"title":"AgentRM: Enhancing Agent Generalization with Reward Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jingru Fan, Maosong Sun, Siyu Yan, Weize Chen, Xin Cong, Yankai Lin, Yaxi Lu, Yu Xia, Zhiyuan Liu, Zhong Zhang","submitted_at":"2025-02-25T17:58:02Z","abstract_excerpt":"Existing LLM-based agents have achieved strong performance on held-in tasks, but their generalizability to unseen tasks remains poor. Hence, some recent work focus on fine-tuning the policy model with more diverse tasks to improve the generalizability. In this work, we find that finetuning a reward model to guide the policy model is more robust than directly finetuning the policy model. Based on this finding, we propose AgentRM, a generalizable reward model, to guide the policy model for effective test-time search. We comprehensively investigate three approaches to construct the reward model, "},"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":"2502.18407","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-25T17:58:02Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"d33ff6f88be2fb81a07880dc6dfbb8f6f4d758ccf40e503078120713955c19fb","abstract_canon_sha256":"d9de390264ae001973350a5bc7dfc249a19539e73d028703049d746619ea4ce1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:19:53.480127Z","signature_b64":"87cgRdNjIkdu7zAv/a3kERQV/apzkZc5oHJ7su6twv8+6JFZxal9FN2NyHeikdhxt+g4bjzwwl8mG3Ux/joCBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ebe9a0f4581efd4342e5894ba02b9c95e76f59c96d3a5b4dd24bcfe05808b8ec","last_reissued_at":"2026-07-05T10:19:53.479660Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:19:53.479660Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AgentRM: Enhancing Agent Generalization with Reward Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jingru Fan, Maosong Sun, Siyu Yan, Weize Chen, Xin Cong, Yankai Lin, Yaxi Lu, Yu Xia, Zhiyuan Liu, Zhong Zhang","submitted_at":"2025-02-25T17:58:02Z","abstract_excerpt":"Existing LLM-based agents have achieved strong performance on held-in tasks, but their generalizability to unseen tasks remains poor. Hence, some recent work focus on fine-tuning the policy model with more diverse tasks to improve the generalizability. In this work, we find that finetuning a reward model to guide the policy model is more robust than directly finetuning the policy model. Based on this finding, we propose AgentRM, a generalizable reward model, to guide the policy model for effective test-time search. We comprehensively investigate three approaches to construct the reward model, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.18407","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/2502.18407/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":"2502.18407","created_at":"2026-07-05T10:19:53.479717+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.18407v1","created_at":"2026-07-05T10:19:53.479717+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.18407","created_at":"2026-07-05T10:19:53.479717+00:00"},{"alias_kind":"pith_short_12","alias_value":"5PU2B5CYD36U","created_at":"2026-07-05T10:19:53.479717+00:00"},{"alias_kind":"pith_short_16","alias_value":"5PU2B5CYD36UGQXF","created_at":"2026-07-05T10:19:53.479717+00:00"},{"alias_kind":"pith_short_8","alias_value":"5PU2B5CY","created_at":"2026-07-05T10:19:53.479717+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25852","citing_title":"Semantic Consistency Policy Optimization for Reinforcement Learning of LLM Agents","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2602.05353","citing_title":"AgentXRay: White-Boxing Agentic Systems via Workflow Reconstruction","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX","json":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX.json","graph_json":"https://pith.science/api/pith-number/5PU2B5CYD36UGQXFRFF2AK44SX/graph.json","events_json":"https://pith.science/api/pith-number/5PU2B5CYD36UGQXFRFF2AK44SX/events.json","paper":"https://pith.science/paper/5PU2B5CY"},"agent_actions":{"view_html":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX","download_json":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX.json","view_paper":"https://pith.science/paper/5PU2B5CY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.18407&json=true","fetch_graph":"https://pith.science/api/pith-number/5PU2B5CYD36UGQXFRFF2AK44SX/graph.json","fetch_events":"https://pith.science/api/pith-number/5PU2B5CYD36UGQXFRFF2AK44SX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX/action/storage_attestation","attest_author":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX/action/author_attestation","sign_citation":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX/action/citation_signature","submit_replication":"https://pith.science/pith/5PU2B5CYD36UGQXFRFF2AK44SX/action/replication_record"}},"created_at":"2026-07-05T10:19:53.479717+00:00","updated_at":"2026-07-05T10:19:53.479717+00:00"}