{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CBKCSWUM2W3OLUVBLRGAD7P4ND","short_pith_number":"pith:CBKCSWUM","schema_version":"1.0","canonical_sha256":"1054295a8cd5b6e5d2a15c4c01fdfc68c9ed59189346dc41310fbdfbf041426c","source":{"kind":"arxiv","id":"2506.20639","version":2},"attestation_state":"computed","paper":{"title":"DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huangjie Zheng, Jiatao Gu, Lingpeng Kong, Navdeep Jaitly, Ruixiang Zhang, Shansan Gong, Yizhe Zhang","submitted_at":"2025-06-25T17:35:47Z","abstract_excerpt":"Diffusion large language models (dLLMs) are compelling alternatives to autoregressive (AR) models because their denoising models operate over the entire sequence. The global planning and iterative refinement features of dLLMs are particularly useful for code generation. However, current training and inference mechanisms for dLLMs in coding are still under-explored. To demystify the decoding behavior of dLLMs and unlock their potential for coding, we systematically investigate their denoising processes and reinforcement learning (RL) methods. We train a 7B dLLM, \\textbf{DiffuCoder}, on 130B tok"},"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":"2506.20639","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-25T17:35:47Z","cross_cats_sorted":[],"title_canon_sha256":"12c3bf5233386fd910b277158330abfe0f74d4c3df0d4d562aabd361d249a508","abstract_canon_sha256":"23d55999f610997c3328e403305a6560c066bf9e983590f3cf05c8402edd8cf8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:32.122085Z","signature_b64":"8ELKAFUnwhLjrEebr65PTF4j6lqaPdNs40xfBK3j8KArQbH7hqqA84LUghD+1WWY0wMPhSqtJpt62zS0nK2qCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1054295a8cd5b6e5d2a15c4c01fdfc68c9ed59189346dc41310fbdfbf041426c","last_reissued_at":"2026-07-05T11:27:32.121619Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:32.121619Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huangjie Zheng, Jiatao Gu, Lingpeng Kong, Navdeep Jaitly, Ruixiang Zhang, Shansan Gong, Yizhe Zhang","submitted_at":"2025-06-25T17:35:47Z","abstract_excerpt":"Diffusion large language models (dLLMs) are compelling alternatives to autoregressive (AR) models because their denoising models operate over the entire sequence. The global planning and iterative refinement features of dLLMs are particularly useful for code generation. However, current training and inference mechanisms for dLLMs in coding are still under-explored. To demystify the decoding behavior of dLLMs and unlock their potential for coding, we systematically investigate their denoising processes and reinforcement learning (RL) methods. We train a 7B dLLM, \\textbf{DiffuCoder}, on 130B tok"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.20639","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/2506.20639/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":"2506.20639","created_at":"2026-07-05T11:27:32.121682+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.20639v2","created_at":"2026-07-05T11:27:32.121682+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.20639","created_at":"2026-07-05T11:27:32.121682+00:00"},{"alias_kind":"pith_short_12","alias_value":"CBKCSWUM2W3O","created_at":"2026-07-05T11:27:32.121682+00:00"},{"alias_kind":"pith_short_16","alias_value":"CBKCSWUM2W3OLUVB","created_at":"2026-07-05T11:27:32.121682+00:00"},{"alias_kind":"pith_short_8","alias_value":"CBKCSWUM","created_at":"2026-07-05T11:27:32.121682+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":38,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05722","citing_title":"Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding","ref_index":55,"is_internal_anchor":true},{"citing_arxiv_id":"2606.18195","citing_title":"Learning from the Self-future: On-policy Self-distillation for dLLMs","ref_index":34,"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":24,"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":22,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00208","citing_title":"SLIM-RL: Risk-Budgeted Random-Masking RL for Diffusion LLMs Without Trajectory Slicing","ref_index":7,"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":55,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04027","citing_title":"MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17842","citing_title":"SNLP: Layer-Parallel Inference via Structured Newton Corrections","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25638","citing_title":"Reinforcement Learning from Denoising Feedback","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29123","citing_title":"The Confidence Shortcut: A Reasoning Failure Mode of Masked Diffusion Models","ref_index":5,"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":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30553","citing_title":"Destruction is a General Strategy to Learn Generation; Diffusion's Strength is to Take it Seriously; Exploration is the Future","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30876","citing_title":"dMoE: dLLMs with Learnable Block Experts","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00724","citing_title":"WaveFilter: Enhancing the Long-Context Capability of Diffusion LLMs via Wavelet-Guided KV Cache Filtering","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06031","citing_title":"NAVIRA: Decoupled Stochastic Remasking for Masked Diffusion Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2511.18801","citing_title":"PartDiffuser: Part-wise 3D Mesh Generation via Discrete Diffusion","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2602.06462","citing_title":"Diffusion-State Policy Optimization for Masked Diffusion Language Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08659","citing_title":"Pushing Biomolecular Utility-Diversity Frontiers with Supergroup Relative Policy Optimization","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16342","citing_title":"DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language Models","ref_index":39,"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":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17842","citing_title":"SNLP: Layer-Parallel Inference via Structured Newton Corrections","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16829","citing_title":"Constrained Code Generation with Discrete Diffusion","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2509.20624","citing_title":"FS-DFM: Fast and Accurate Long Text Generation with Few-Step Diffusion Language Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2510.03206","citing_title":"Coevolutionary Continuous Discrete Diffusion: Make Your Diffusion Language Model a Latent Reasoner","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND","json":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND.json","graph_json":"https://pith.science/api/pith-number/CBKCSWUM2W3OLUVBLRGAD7P4ND/graph.json","events_json":"https://pith.science/api/pith-number/CBKCSWUM2W3OLUVBLRGAD7P4ND/events.json","paper":"https://pith.science/paper/CBKCSWUM"},"agent_actions":{"view_html":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND","download_json":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND.json","view_paper":"https://pith.science/paper/CBKCSWUM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.20639&json=true","fetch_graph":"https://pith.science/api/pith-number/CBKCSWUM2W3OLUVBLRGAD7P4ND/graph.json","fetch_events":"https://pith.science/api/pith-number/CBKCSWUM2W3OLUVBLRGAD7P4ND/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND/action/storage_attestation","attest_author":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND/action/author_attestation","sign_citation":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND/action/citation_signature","submit_replication":"https://pith.science/pith/CBKCSWUM2W3OLUVBLRGAD7P4ND/action/replication_record"}},"created_at":"2026-07-05T11:27:32.121682+00:00","updated_at":"2026-07-05T11:27:32.121682+00:00"}