{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OWBKCNRABECSZF3SSBVN76TN5V","short_pith_number":"pith:OWBKCNRA","schema_version":"1.0","canonical_sha256":"7582a1362009052c9772906adffa6ded452cb32fcaffc89d96bf07d1e624f7bb","source":{"kind":"arxiv","id":"2509.06949","version":1},"attestation_state":"computed","paper":{"title":"Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bowen Li, Ke Shen, Ling Yang, Mengdi Wang, Ye Tian, Yinjie Wang","submitted_at":"2025-09-08T17:58:06Z","abstract_excerpt":"We propose TraceRL, a trajectory-aware reinforcement learning framework for diffusion language models (DLMs) that incorporates preferred inference trajectory into post-training, and is applicable across different architectures. Equipped with a diffusion-based value model that enhances training stability, we demonstrate improved reasoning performance on complex math and coding tasks. Besides, it can also be applied to adapt block-specific models to larger blocks, which improves sampling flexibility. Employing TraceRL, we derive a series of state-of-the-art diffusion language models, namely TraD"},"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":"2509.06949","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-08T17:58:06Z","cross_cats_sorted":[],"title_canon_sha256":"2a7fe0efb33f8c143d1399b7c4aa9d58bc9909fb0841131ce6e91556968118c4","abstract_canon_sha256":"be1c44c753ac1d6624a1ebccbc0fb4c7ef79b50898d591b6401fe51189ba152d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:06:55.903488Z","signature_b64":"QZN4MMSgqYVo883s+3RfXCpoROyFt4P7FLvIRwBk+cWvcYB01EsJWrG2GfuxWCajliy/XhUSnOosrTjip7g9Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7582a1362009052c9772906adffa6ded452cb32fcaffc89d96bf07d1e624f7bb","last_reissued_at":"2026-07-05T12:06:55.902871Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:06:55.902871Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bowen Li, Ke Shen, Ling Yang, Mengdi Wang, Ye Tian, Yinjie Wang","submitted_at":"2025-09-08T17:58:06Z","abstract_excerpt":"We propose TraceRL, a trajectory-aware reinforcement learning framework for diffusion language models (DLMs) that incorporates preferred inference trajectory into post-training, and is applicable across different architectures. Equipped with a diffusion-based value model that enhances training stability, we demonstrate improved reasoning performance on complex math and coding tasks. Besides, it can also be applied to adapt block-specific models to larger blocks, which improves sampling flexibility. Employing TraceRL, we derive a series of state-of-the-art diffusion language models, namely TraD"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.06949","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/2509.06949/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":"2509.06949","created_at":"2026-07-05T12:06:55.902942+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.06949v1","created_at":"2026-07-05T12:06:55.902942+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.06949","created_at":"2026-07-05T12:06:55.902942+00:00"},{"alias_kind":"pith_short_12","alias_value":"OWBKCNRABECS","created_at":"2026-07-05T12:06:55.902942+00:00"},{"alias_kind":"pith_short_16","alias_value":"OWBKCNRABECSZF3S","created_at":"2026-07-05T12:06:55.902942+00:00"},{"alias_kind":"pith_short_8","alias_value":"OWBKCNRA","created_at":"2026-07-05T12:06:55.902942+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21633","citing_title":"HERALD: High-Throughput Block Diffusion LLM Serving via CPU-GPU Cooperative KV Cache Retrieval","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2606.17999","citing_title":"VoidPadding: Let [VOID] Handle Padding in Masked Diffusion Language Models so that [EOS] Can Focus on Semantic Termination","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18195","citing_title":"Learning from the Self-future: On-policy Self-distillation for dLLMs","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01893","citing_title":"Spec-AUF: Accept-Until-Fail Training under Train-Inference Misalignment for Masked Block Drafters","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12273","citing_title":"Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09159","citing_title":"Unified Energy for Invariant and Independent Decoding in Diffusion Language Models","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08501","citing_title":"Back on Track: Aligning Rewards and States for Reasoning in Diffusion Large Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06712","citing_title":"Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04535","citing_title":"Dynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01170","citing_title":"Diffusion-GR2: Diffusion Generative Reasoning Re-ranker","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23163","citing_title":"Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29398","citing_title":"GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18176","citing_title":"Improving Sampling for Masked Diffusion Models via Information Gain","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23163","citing_title":"Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17174","citing_title":"Beyond Execution: Static-Analysis Rewards and Hint-Conditioned Diffusion RL for Code Generation","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16842","citing_title":"Sketch Then Paint: Hierarchical Reinforcement Learning for Diffusion Multi-Modal Large Language Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13935","citing_title":"Beyond Mode-Seeking RL: Trajectory-Balance Post-Training for Diffusion Language Models","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2603.22241","citing_title":"MemDLM: Memory-Enhanced DLM Training","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10218","citing_title":"Relative Score Policy Optimization for Diffusion Language Models","ref_index":95,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09536","citing_title":"TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V","json":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V.json","graph_json":"https://pith.science/api/pith-number/OWBKCNRABECSZF3SSBVN76TN5V/graph.json","events_json":"https://pith.science/api/pith-number/OWBKCNRABECSZF3SSBVN76TN5V/events.json","paper":"https://pith.science/paper/OWBKCNRA"},"agent_actions":{"view_html":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V","download_json":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V.json","view_paper":"https://pith.science/paper/OWBKCNRA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.06949&json=true","fetch_graph":"https://pith.science/api/pith-number/OWBKCNRABECSZF3SSBVN76TN5V/graph.json","fetch_events":"https://pith.science/api/pith-number/OWBKCNRABECSZF3SSBVN76TN5V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V/action/storage_attestation","attest_author":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V/action/author_attestation","sign_citation":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V/action/citation_signature","submit_replication":"https://pith.science/pith/OWBKCNRABECSZF3SSBVN76TN5V/action/replication_record"}},"created_at":"2026-07-05T12:06:55.902942+00:00","updated_at":"2026-07-05T12:06:55.902942+00:00"}