{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YH2KS66NZI4KHPEIHYL7A54ZJ2","short_pith_number":"pith:YH2KS66N","schema_version":"1.0","canonical_sha256":"c1f4a97bcdca38a3bc883e17f077994ebfe6e73603cdb27af60cbf28a2cffd62","source":{"kind":"arxiv","id":"2506.02208","version":1},"attestation_state":"computed","paper":{"title":"KDRL: Post-Training Reasoning LLMs via Unified Knowledge Distillation and Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Fei Mi, Heyuan Deng, Hongling Xu, Jinpeng Li, Lifeng Shang, Lu Hou, Qi Zhu, Ruifeng Xu, Yasheng Wang","submitted_at":"2025-06-02T19:46:41Z","abstract_excerpt":"Recent advances in large language model (LLM) post-training have leveraged two distinct paradigms to enhance reasoning capabilities: reinforcement learning (RL) and knowledge distillation (KD). While RL enables the emergence of complex reasoning behaviors, it often suffers from low sample efficiency when the initial policy struggles to explore high-reward trajectories. Conversely, KD improves learning efficiency via mimicking the teacher model but tends to generalize poorly to out-of-domain scenarios. In this work, we present \\textbf{KDRL}, a \\textit{unified post-training framework} that joint"},"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":"2506.02208","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-02T19:46:41Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"9a9c05406029c353ddde54563c5f6b983b55028d907667d91d459d51067430d6","abstract_canon_sha256":"28638f0d4c8cb094725f000140dd681200b425db203d1c1600c154343221046d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:34.392272Z","signature_b64":"JwsLdqtzLK7QvVZYeAfSrofoQIUq3fsUBkv5rLkC3QETH5QG+oYKuX0L6spVHQUjAH7J4KqUIAWJ1Horr6O+Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1f4a97bcdca38a3bc883e17f077994ebfe6e73603cdb27af60cbf28a2cffd62","last_reissued_at":"2026-07-05T11:14:34.391770Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:34.391770Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"KDRL: Post-Training Reasoning LLMs via Unified Knowledge Distillation and Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Fei Mi, Heyuan Deng, Hongling Xu, Jinpeng Li, Lifeng Shang, Lu Hou, Qi Zhu, Ruifeng Xu, Yasheng Wang","submitted_at":"2025-06-02T19:46:41Z","abstract_excerpt":"Recent advances in large language model (LLM) post-training have leveraged two distinct paradigms to enhance reasoning capabilities: reinforcement learning (RL) and knowledge distillation (KD). While RL enables the emergence of complex reasoning behaviors, it often suffers from low sample efficiency when the initial policy struggles to explore high-reward trajectories. Conversely, KD improves learning efficiency via mimicking the teacher model but tends to generalize poorly to out-of-domain scenarios. In this work, we present \\textbf{KDRL}, a \\textit{unified post-training framework} that joint"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.02208","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/2506.02208/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":"2506.02208","created_at":"2026-07-05T11:14:34.391828+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.02208v1","created_at":"2026-07-05T11:14:34.391828+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.02208","created_at":"2026-07-05T11:14:34.391828+00:00"},{"alias_kind":"pith_short_12","alias_value":"YH2KS66NZI4K","created_at":"2026-07-05T11:14:34.391828+00:00"},{"alias_kind":"pith_short_16","alias_value":"YH2KS66NZI4KHPEI","created_at":"2026-07-05T11:14:34.391828+00:00"},{"alias_kind":"pith_short_8","alias_value":"YH2KS66N","created_at":"2026-07-05T11:14:34.391828+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05394","citing_title":"Weak-to-Strong Generalization via Direct On-Policy Distillation","ref_index":63,"is_internal_anchor":true},{"citing_arxiv_id":"2606.18216","citing_title":"Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01464","citing_title":"Cross-lingual Self-Consistency for Multilingual Reasoning with Language Models","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22156","citing_title":"One-Way Policy Optimization for Self-Evolving LLMs","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08401","citing_title":"AIPO: Learning to Reason from Active Interaction","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12652","citing_title":"Multi-Rollout On-Policy Distillation via Peer Successes and Failures","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13230","citing_title":"Teacher-Guided Policy Optimization for On-Policy Reasoning Distillation under Large Policy Divergence","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08873","citing_title":"CoDistill-GRPO: A Co-Distillation Recipe for Efficient Group Relative Policy Optimization","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08401","citing_title":"AIPO: Learning to Reason from Active Interaction","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10674","citing_title":"Skill-SD: Skill-Conditioned Self-Distillation for Multi-turn LLM Agents","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07276","citing_title":"Signal Reshaping for GRPO in Weak-Feedback Agentic Code Repair","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07274","citing_title":"Structured Role-Aware Policy Optimization for Multimodal Reasoning","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17614","citing_title":"Characterizing Model-Native Skills","ref_index":86,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2","json":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2.json","graph_json":"https://pith.science/api/pith-number/YH2KS66NZI4KHPEIHYL7A54ZJ2/graph.json","events_json":"https://pith.science/api/pith-number/YH2KS66NZI4KHPEIHYL7A54ZJ2/events.json","paper":"https://pith.science/paper/YH2KS66N"},"agent_actions":{"view_html":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2","download_json":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2.json","view_paper":"https://pith.science/paper/YH2KS66N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.02208&json=true","fetch_graph":"https://pith.science/api/pith-number/YH2KS66NZI4KHPEIHYL7A54ZJ2/graph.json","fetch_events":"https://pith.science/api/pith-number/YH2KS66NZI4KHPEIHYL7A54ZJ2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2/action/storage_attestation","attest_author":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2/action/author_attestation","sign_citation":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2/action/citation_signature","submit_replication":"https://pith.science/pith/YH2KS66NZI4KHPEIHYL7A54ZJ2/action/replication_record"}},"created_at":"2026-07-05T11:14:34.391828+00:00","updated_at":"2026-07-05T11:14:34.391828+00:00"}