{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7TVFGYSUE73UM2EGIXPOH26OL5","short_pith_number":"pith:7TVFGYSU","schema_version":"1.0","canonical_sha256":"fcea53625427f746688645dee3ebce5f6d1f7b2dafe8968854e5f1dcb38b2982","source":{"kind":"arxiv","id":"2505.15692","version":5},"attestation_state":"computed","paper":{"title":"TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chonghua Liao, Haoran Luo, Huazhe Xu, Jianhua Tao, Jinyang Wu, Ling Yang, Mingkuan Feng, Shuai Zhang, Zhengqi Wen","submitted_at":"2025-05-21T16:06:10Z","abstract_excerpt":"Reinforcement learning (RL) has emerged as an effective paradigm for enhancing model reasoning. However, existing RL methods like GRPO typically rely on unstructured self-sampling to fit scalar rewards, often producing inefficient rollouts that fail to capture transferable problem-solving strategies. To address this limitation, we propose **TemplateRL**, a structured template-guided RL framework that augments policy optimization with explicit template guidance. Our approach first constructs a problem-solving template library via MCTS on a small seed set, then seamlessly integrates this high-le"},"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":"2505.15692","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-21T16:06:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"116ddaa3dcc5fa075df3a4d73a82a493286017b72d81a2d833da7df23115b7b1","abstract_canon_sha256":"562870bc9c6d93f8909fff6501d037b6c81241a9e8f78569afdae24d65cdc70e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:20.137109Z","signature_b64":"TF7t61JVlTfAxLZ8KE7q743ecVQehOUa08IDitDtDmTuvV/k9oiRDwEq0+zynMaAdLYoDuWVANpGBM6HcTmNAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fcea53625427f746688645dee3ebce5f6d1f7b2dafe8968854e5f1dcb38b2982","last_reissued_at":"2026-05-20T00:00:20.136359Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:20.136359Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chonghua Liao, Haoran Luo, Huazhe Xu, Jianhua Tao, Jinyang Wu, Ling Yang, Mingkuan Feng, Shuai Zhang, Zhengqi Wen","submitted_at":"2025-05-21T16:06:10Z","abstract_excerpt":"Reinforcement learning (RL) has emerged as an effective paradigm for enhancing model reasoning. However, existing RL methods like GRPO typically rely on unstructured self-sampling to fit scalar rewards, often producing inefficient rollouts that fail to capture transferable problem-solving strategies. To address this limitation, we propose **TemplateRL**, a structured template-guided RL framework that augments policy optimization with explicit template guidance. Our approach first constructs a problem-solving template library via MCTS on a small seed set, then seamlessly integrates this high-le"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15692","kind":"arxiv","version":5},"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/2505.15692/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":"2505.15692","created_at":"2026-05-20T00:00:20.136491+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15692v5","created_at":"2026-05-20T00:00:20.136491+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15692","created_at":"2026-05-20T00:00:20.136491+00:00"},{"alias_kind":"pith_short_12","alias_value":"7TVFGYSUE73U","created_at":"2026-05-20T00:00:20.136491+00:00"},{"alias_kind":"pith_short_16","alias_value":"7TVFGYSUE73UM2EG","created_at":"2026-05-20T00:00:20.136491+00:00"},{"alias_kind":"pith_short_8","alias_value":"7TVFGYSU","created_at":"2026-05-20T00:00:20.136491+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2508.07809","citing_title":"EvoCoT: Overcoming the Exploration Bottleneck in Reinforcement Learning","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2604.23747","citing_title":"SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00610","citing_title":"Decouple before Integration: Test-time Synthesis of SFT and RLVR Task Vectors","ref_index":46,"is_internal_anchor":true},{"citing_arxiv_id":"2604.07941","citing_title":"Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning","ref_index":90,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5","json":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5.json","graph_json":"https://pith.science/api/pith-number/7TVFGYSUE73UM2EGIXPOH26OL5/graph.json","events_json":"https://pith.science/api/pith-number/7TVFGYSUE73UM2EGIXPOH26OL5/events.json","paper":"https://pith.science/paper/7TVFGYSU"},"agent_actions":{"view_html":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5","download_json":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5.json","view_paper":"https://pith.science/paper/7TVFGYSU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15692&json=true","fetch_graph":"https://pith.science/api/pith-number/7TVFGYSUE73UM2EGIXPOH26OL5/graph.json","fetch_events":"https://pith.science/api/pith-number/7TVFGYSUE73UM2EGIXPOH26OL5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5/action/storage_attestation","attest_author":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5/action/author_attestation","sign_citation":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5/action/citation_signature","submit_replication":"https://pith.science/pith/7TVFGYSUE73UM2EGIXPOH26OL5/action/replication_record"}},"created_at":"2026-05-20T00:00:20.136491+00:00","updated_at":"2026-05-20T00:00:20.136491+00:00"}